In the communication cycle, changing the idea into words, pictures or gestures is called
(b) encoding Encoding turns the sender's idea into a message (words, pictures, gestures); decoding is the receiver understanding it.
Minimum Learning Support Material · CBSE 2026-27
Notes · Numericals with working · 684 solved questions · Flashcards · Quiz · Practice Paper Maker
CBSE Sample Question Paper 2026-27: 21 questions, 50 marks, 2 hours
| Question | What comes | Attempt | Marks |
|---|---|---|---|
| Q1 | 6 objective questions on Employability Skills | any 4 | 1 x 4 = 4 |
| Q2 to Q5 | 6 objective questions each on the AI units (MCQ, True/False statements, Assertion-Reason, small numericals) | any 5 in each | 4 x 5 = 20 |
| Q6 to Q10 | Employability Skills, 20 to 30 words each | any 3 of 5 | 2 x 3 = 6 |
| Q11 to Q16 | AI units, 20 to 30 words each | any 4 of 6 | 2 x 4 = 8 |
| Q17 to Q21 | AI units, 50 to 80 words each: case study, neural network, confusion matrix numerical, Bag of Words numerical | any 3 of 5 | 4 x 3 = 12 |
The paper has 21 questions and is of 2 hours. You answer 15 questions: all of Q1 to Q5 (with the choice inside each) and 10 out of Q6 to Q21.
Unit 4 (Statistical Data) and Unit 7 (Advance Python) are assessed in the practical only.
One common blueprint, as per the CBSE Sample Question Paper 2026-27. All the practice papers of this kit follow it. Each cell shows the number of questions given from that unit.
| Unit | 1mark | 2marks | 4marks | Totalquestions |
|---|---|---|---|---|
| Part ACommunication Skills-II | 1 | 1 | - | 2 |
| Part ASelf-Management Skills-II | 2 | 1 | - | 3 |
| Part AICT Skills-II | 1 | 1 | - | 2 |
| Part AEntrepreneurial Skills-II | 1 | 1 | - | 2 |
| Part AGreen Skills-II | 1 | 1 | - | 2 |
| Part BUnit 1: AI Project Cycle and Ethical Frameworks | 5 | 1 | 1 | 7 |
| Part BUnit 2: Advanced Concepts of Modeling in AI | 4 | 2 | 2 | 8 |
| Part BUnit 3: Evaluating Models | 6 | 1 | 1 | 8 |
| Part BUnit 5: Computer Vision | 4 | 1 | - | 5 |
| Part BUnit 6: Natural Language Processing | 5 | 1 | 1 | 7 |
| Questions given | 30 | 11 | 5 | 46 |
| To be answered | 24 | 7 | 3 | 34 |
| Marks | 24 | 14 | 12 | 50 |
Where the 50 marks come from, easiest first
10 marks · five units · detailed chapters
Communication is the process of exchanging information, ideas, thoughts and feelings between people through speech, writing, signs or behaviour, so that both reach the same understanding. In Class IX we learnt the basic cycle; in Class X we study the methods, the role of feedback, the barriers and the writing skills needed at the workplace.
a supervisor explains a task to the team
a new employee asks how to use a machine
sending a daily sales report by e-mail
listening to and solving complaints
showing a new idea in a meeting
sharing updates so that work is not repeated
| Stage | Who / what | Example: a sales manager |
|---|---|---|
| Sender | starts the message | the sales manager |
| Encoding | puts the idea into words or symbols | writes "Target: 50 units this week" |
| Channel | the medium | e-mail to the team |
| Receiver | gets the message | the sales team |
| Decoding | understands the meaning | team understands the weekly target |
| Feedback | response | a member replies, "Does it include online orders?" |
| Oral | Written |
|---|---|
| face-to-face talks, phone calls, meetings, video calls, speeches | letters, e-mails, SMS, notices, reports, minutes |
| fast; immediate feedback; tone helps meaning | permanent record; can be read again; reaches many |
| no record; may be forgotten | slow feedback; needs good language skills |
Tips for effective verbal communication: know your purpose and audience, use simple words, speak clearly at a moderate speed, be polite, avoid slang at work, and check understanding by asking questions.
Non-verbal messages are sent through facial expressions, gestures, posture, eye contact, touch, space and paralanguage (tone, pitch, volume and speed of voice). Non-verbal signals can support, replace or even contradict words. If someone says "I am fine" with tears in the eyes, we believe the non-verbal signal.
| Positive body language | Negative body language |
|---|---|
| smiling, nodding, open palms | frowning, rolling eyes |
| straight posture, leaning slightly forward | slouching, crossed arms |
| steady eye contact | looking away, checking the phone |
| firm handshake | limp handshake or crushing grip |
Visual communication uses pictures, signs, symbols, maps, charts, graphs, infographics and videos. Road signs, safety signs in factories, emojis and company logos are visual. It is quick, crosses language barriers and is remembered easily, but cannot express every detail.
Feedback is the response of the receiver to the sender's message. It completes the communication cycle and tells the sender whether the message was received, understood and accepted.
praise for good work: "Your report was very clear."
points out what needs to change
gives details: "Add the chart on page 2."
vague: "Good" or "Bad" - less useful
Qualities of good feedback: specific, timely, polite, focused on the work and not the person, balanced (good points and improvements) and actionable.
one clear purpose
no unnecessary words
specific facts and figures
right language and facts
logical and connected
all information included
polite and respectful
| Barrier | Example | Solution |
|---|---|---|
| Physical | noisy machines, poor phone network, distance | quiet place, better equipment |
| Linguistic | a customer speaks only Tamil; technical jargon | simple words, translator, common language |
| Interpersonal | lack of trust, anger, shyness | build trust, stay calm, listen actively |
| Organisational | many levels, strict rules, no open discussion | clear channels, open-door policy |
| Cultural | different customs and gestures | respect and learn about other cultures |
| Psychological | stress, prejudice, assumptions | keep an open mind, clarify doubts |
A sentence must have a subject and a verb and must give complete meaning. A simple sentence has one subject and one main verb: "The train arrived late."
| Part of speech | Function | Example in a sentence |
|---|---|---|
| Noun | names a person, place, thing or idea | Priya visited Agra. |
| Pronoun | replaces a noun | She liked it. |
| Verb | shows action or state | She took photos. |
| Adjective | describes a noun | the white marble |
| Adverb | describes a verb, adjective or adverb | she walked slowly |
| Preposition | shows relation or position | inside the tomb |
| Conjunction | joins words or sentences | tired but happy |
| Interjection | shows sudden feeling | Wow! |
| Active (doer first) | Passive (action first) |
|---|---|
| The manager signed the letter. | The letter was signed by the manager. |
| Students planted trees. | Trees were planted by students. |
Use the active voice in most workplace writing because it is direct and clear.
Subject: Request for leave on 12 March. Dear Sir, I request leave on 12 March as I have to attend my sister's wedding. I have completed my pending reports. Thank you. Regards, Ravi Kumar, Sales Assistant.
Feedback completes the communication cycle. It tells the sender whether the message was received and understood correctly, helps correct misunderstandings, and improves future communication.
Verbal communication uses spoken or written words, e.g. a phone call and an e-mail. Non-verbal communication sends messages without words through body language and tone, e.g. a smile and a thumbs-up. Verbal messages carry the content; non-verbal signals show feelings and can support or contradict the words.
Self-management is the ability to manage our emotions, thoughts, time and actions to achieve our goals and do well at school and work. In Class X the focus is on stress management, self-awareness, self-motivation, goal setting and time management.
Stress is the feeling of pressure or tension when we face a difficult or demanding situation - exams, competitions, a new school, family problems. A small amount of stress (called eustress) pushes us to perform; too much stress (distress) harms health and performance.
| Physical | Emotional | Behavioural |
|---|---|---|
| headache, tiredness, poor sleep, stomach ache | worry, anger, irritation, sadness | avoiding work, overeating or not eating, too much phone use |
plan and finish work early
games, cycling, walking
deep breathing calms the mind
8 hours of sleep, less junk food
music, gardening, art
share with family, friends, counsellor
focus on what you can control
Adversity (event): Ravi scores low in a unit test. Belief (thought): "I am a failure." Consequence: he feels sad and stops studying. If Ravi changes the belief to "I need more practice in algebra", the consequence becomes extra practice and better marks. Changing our thinking changes how stress affects us.
Self-awareness is understanding our own strengths, weaknesses, values, interests and emotions. It helps us choose the right subjects, careers and friends.
| Know your | Question to ask | Example |
|---|---|---|
| Strengths | What do I do well? | I explain concepts well to friends |
| Weaknesses | What do I find hard? | I get nervous in interviews |
| Interests | What do I enjoy? | coding, cricket |
| Values | What matters to me? | honesty, helping others |
Self-motivation is the inner force that makes us start and continue doing something without being pushed by others.
| Internal (intrinsic) | External (extrinsic) |
|---|---|
| comes from within: interest, satisfaction, curiosity | comes from outside: marks, prizes, money, praise |
| learning to code because it is fun | studying only to get a new phone |
clear goals
learn new skills
take responsibility
keep trying after failure
How to build self-motivation: find your why, set goals, break them into small steps, track progress, reward yourself, and stay with motivated people.
Goal setting means deciding what we want to achieve and planning how to get there.
| Letter | Means | Check question |
|---|---|---|
| S | Specific | Is the goal clear and exact? |
| M | Measurable | Can I count or measure progress? |
| A | Achievable | Is it possible for me? |
| R | Realistic / Relevant | Does it fit my situation and aims? |
| T | Time-bound | Is there a deadline? |
Wish: "I want to be fit." SMART goal: "I will jog for 20 minutes every morning, 5 days a week, and reduce my 1 km running time from 8 minutes to 6 minutes by 31 March."
Time management is planning and controlling how we spend our time to do the most important work efficiently.
| Step | What to do |
|---|---|
| Organise | list all tasks; use a diary, calendar or app |
| Prioritise | do the most important and urgent work first |
| Control | avoid distractions such as social media; say no politely |
| Track | check at the end of the day what was done and what is left |
| Urgent | Not urgent | |
|---|---|---|
| Important | Do now: assignment due tomorrow | Plan: revision for exams next month |
| Not important | Delegate or do quickly: some calls | Avoid: endless scrolling |
Time wasters: procrastination (delaying work), too much phone use, unclear goals, doing too many things at once.
Self-regulation means controlling our behaviour and emotions - not reacting in anger, keeping promises, completing work on time. It includes self-discipline, adaptability and accountability.
Time management - planning work reduces last-minute pressure. Physical exercise and yoga - release tension and improve mood. Healthy diet and enough sleep - keep the body strong. Talking to friends and family - sharing reduces worry. Positive thinking and hobbies also help.
Organise, Prioritise, Control and Track.
An operating system (OS) is the system software that manages the computer's hardware and software and acts as an interface between the user and the computer. Without an OS, we cannot run applications. Examples: Windows, Linux (Ubuntu), macOS, Android, iOS.
runs many programs; shares the CPU
gives RAM to programs and frees it
creates, saves, organises and deletes files
controls keyboard, printer, disk through drivers
passwords and user accounts
lets us interact with the computer
| Type | Example |
|---|---|
| Desktop OS | Windows 11, Ubuntu, macOS |
| Mobile OS | Android, iOS |
| Open source (free to use and change) | Linux, Android (AOSP) |
| Proprietary (owned by a company) | Windows, macOS |
| Operation | How (Windows) |
|---|---|
| Create a folder | right-click > New > Folder |
| Rename | select > F2 or right-click > Rename |
| Copy / Cut / Paste | Ctrl + C / Ctrl + X / Ctrl + V |
| Delete | Delete key; file goes to the Recycle Bin |
| Restore | open the Recycle Bin > Restore |
| Permanently delete | Shift + Delete |
| Search | search box in File Explorer |
A file extension tells the type of a file: .docx (Word), .xlsx (Excel), .pptx (PowerPoint), .pdf, .jpg, .mp4, .py, .txt.
Regular care keeps the computer fast and long-lasting.
wipe the keyboard and screen; keep vents dust-free
install OS and software updates
run an antivirus scan weekly
delete unwanted files, empty the Recycle Bin
copy important data to a drive or cloud
use a UPS and surge protector
Do not eat or drink near the computer, do not move a laptop while the hard disk is working, handle cables carefully and shut down properly.
| Threat | What it does | How it spreads |
|---|---|---|
| Virus | a program that attaches to files and harms data; runs when the file runs | infected files, pen drives, downloads |
| Worm | copies itself and spreads automatically across networks; slows systems | networks, e-mail |
| Trojan horse | looks like a useful program but harms when run; does not copy itself | fake apps, cracked software |
| Spyware | secretly records what you do (passwords, browsing) | free software, malicious sites |
| Adware | shows unwanted advertisements | free apps |
| Ransomware | locks files and demands money | e-mail attachments |
| Phishing | fake e-mails or messages to steal passwords and bank details | e-mail, SMS, fake websites |
| Hacking | unauthorised access to a computer or account | weak passwords, unprotected networks |
Signs of infection: the computer becomes slow, programs crash, unknown pop-ups appear, files disappear or change, settings change by themselves.
| Measure | Explanation |
|---|---|
| Strong password | at least 8 characters with capitals, small letters, numbers and symbols, e.g. Kv@2026#Ai; change regularly |
| Antivirus | detects and removes malware; keep it updated (e.g. Windows Defender, Quick Heal) |
| Firewall | blocks unauthorised network access |
| Updates | fix security holes in the OS and apps |
| Encryption | turns data into a secret code so others cannot read it |
| Backup | keeps a copy safe in case of loss |
| Safe browsing | use https sites, do not click unknown links, do not download pirated software |
| Two-factor authentication | password + OTP for extra safety |
Neha gets an SMS: "Your bank account will be blocked today. Click here and enter your PIN." It is phishing - banks never ask for PINs. She deletes the message and reports it at cybercrime.gov.in (helpline 1930).
A virus attaches itself to a file and spreads when the infected file is run or shared. A worm is a standalone program that copies itself and spreads automatically through networks without user action.
Use strong passwords and change them regularly; install and update antivirus software; turn on the firewall and install updates; avoid unknown links, attachments and pirated software; back up important data regularly.
Entrepreneurship is the process of identifying an opportunity, starting a business, taking risks and managing resources to provide goods or services and earn profit. An entrepreneur is the person who does this.
Entrepreneurs are important for society because they create jobs, bring new products, develop regions, use local resources and increase the country's income. Every big company - Tata, Infosys, Zomato, Paytm - was once a small idea.
Arunachalam Muruganantham from Tamil Nadu made a low-cost machine to produce affordable sanitary pads. Rural women run these machines, earn money and improve health in their villages. He earned profit and also solved a social problem - this is social entrepreneurship.
success takes time
sees the good in problems
long hours in the start-up phase
never gives up after failure
believes in the idea
learns from mistakes
finds new solutions
builds and guides a team
| Function | Explanation | Example: a bakery |
|---|---|---|
| Decision making | what to produce, where, at what price | make eggless cakes for the local market |
| Managing the business | planning, hiring, controlling daily work | hire two bakers; plan stock |
| Dividing income | paying wages, rent, suppliers, keeping profit | pay salaries before keeping profit |
| Taking risk | bearing the chance of loss | buys an expensive oven on loan |
| Creating new methods (innovation) | improving products and processes | online ordering through a website |
| Myth | Fact |
|---|---|
| Starting a business needs a lot of money | many start with little money (home tiffin, tuition) and grow |
| Entrepreneurs are born, not made | skills can be learnt by training and practice |
| Only big ideas succeed | small improvements to existing products also succeed |
| Failure means the end | many entrepreneurs fail first and then succeed |
| A business is for older people only | many students start online businesses and startups |
| Stage | Key activity |
|---|---|
| Self-analysis | know your interests, skills and capacity for risk |
| Opportunity | find unmet needs: market survey, observation |
| Idea | brainstorm ideas, choose the most practical one |
| Business plan | product, customers, cost, price, marketing, profit estimate |
| Finance | own savings, family, bank loan, government schemes |
| Launch | start small and test |
| Growth | add products, branches, online sales |
Startup India, Stand-Up India (loans for SC/ST and women entrepreneurs), PM Mudra Yojana (Shishu, Kishor, Tarun loans for small businesses), Skill India and Atal Innovation Mission (Atal Tinkering Labs in schools).
| Type | Example |
|---|---|
| Manufacturing | furniture making, textile mill |
| Trading | grocery store, wholesale market |
| Service | salon, coaching, courier |
| Agriculture-based | dairy, poultry, organic farm |
| Online / e-commerce | selling handicrafts online |
A tailor buys cloth worth Rs 6,000 and spends Rs 2,000 on thread and electricity. She sells shirts for Rs 12,500. Profit = 12,500 - (6,000 + 2,000) = Rs 4,500.
In 1946, dairy farmers of Kheda (Gujarat) were paid very little by middlemen. Under the leadership of Tribhuvandas Patel and later Dr Verghese Kurien, they formed a cooperative. Farmers sold milk directly, processing plants were set up and the brand Amul was born. Today millions of farmers are members, and India became the world's largest milk producer (White Revolution). Lessons: spot a real problem, organise people, maintain quality, and keep the customer first.
solve a real need
quality and honesty create loyal customers
test, learn, then expand
feedback shapes the product
| Question | Plan section |
|---|---|
| What will I sell? | product or service |
| Who will buy it? | target customers |
| How much will it cost and at what price will I sell? | costs and pricing |
| How will people know about it? | promotion: posters, social media, word of mouth |
| Where will I sell? | place: shop, online, door to door |
| How much money do I need? | capital and sources |
Decision making - deciding what to produce, how and at what price. Risk taking - bearing the possibility of loss when investing money and effort. (Others: managing, dividing income, innovation.)
Myth: it needs a lot of money - fact: many start small. Myth: entrepreneurs are born - fact: skills can be learnt. Myth: only new ideas succeed - fact: improving old ideas works. Myth: failure is the end - fact: failure teaches and many succeed later.
Sustainable development is development that meets the needs of the present generation without compromising the ability of future generations to meet their own needs (Brundtland Report, 1987). It means using resources wisely, so that our children also get clean air, water, forests and minerals.
soil loss, overuse of fertilisers, food waste
over-pumping of groundwater, polluted rivers
dependence on coal and oil
deforestation and forest fires
crowding, waste and traffic
global warming, floods and droughts
| Problem | Cause | Sustainable solution |
|---|---|---|
| Soil degradation | chemical fertilisers, overgrazing | organic farming, crop rotation |
| Water shortage | wasting water, deforestation | rainwater harvesting, drip irrigation |
| Energy crisis | fossil fuels are limited | solar, wind, hydro, biogas |
| Air pollution | vehicles, industry, crop burning | public transport, EVs, cleaner fuels |
| Waste | plastic, e-waste | reduce, reuse, recycle, segregate |
In 2015, the United Nations adopted 17 Sustainable Development Goals to be achieved by 2030.
| No. | Goal | No. | Goal |
|---|---|---|---|
| 1 | No poverty | 10 | Reduced inequalities |
| 2 | Zero hunger | 11 | Sustainable cities and communities |
| 3 | Good health and well-being | 12 | Responsible consumption and production |
| 4 | Quality education | 13 | Climate action |
| 5 | Gender equality | 14 | Life below water |
| 6 | Clean water and sanitation | 15 | Life on land |
| 7 | Affordable and clean energy | 16 | Peace, justice and strong institutions |
| 8 | Decent work and economic growth | 17 | Partnerships for the goals |
| 9 | Industry, innovation and infrastructure |
A school (SDG 4) teaches girls coding (SDG 5), runs on solar power (SDG 7) and harvests rainwater (SDG 6). A single project can support many goals.
A green economy is low-carbon, resource-efficient and socially inclusive. Green jobs are jobs that help protect or restore the environment.
installs and repairs solar panels
services electric vehicles
sorts and processes waste
grows crops without chemicals
energy-saving buildings
rainwater harvesting systems
| At home | At school | In society |
|---|---|---|
| turn off taps, lights and fans | eco-club, composting pit | cleanliness and plantation drives |
| use cloth bags | use both sides of paper | use public transport |
| repair before replacing | no food waste in the canteen | report pollution, support local products |
India co-founded it to promote solar power worldwide
Lifestyle for Environment - mindful use of resources
tap water to every rural home
sanitation and waste management
electric vehicles and charging network
increase forest cover
Hiware Bazar in Maharashtra once suffered from drought and migration. The villagers built check dams, planted trees, banned water-hungry crops and managed groundwater together. Within a few years, wells were full, farm incomes rose and people returned to the village. It shows that community action can make development sustainable.
| Area | Sustainable choice |
|---|---|
| Food | eat local and seasonal food; do not waste food |
| Water | bucket bath; reuse RO waste water for plants |
| Energy | 5-star appliances; switch off at the plug |
| Travel | walk, cycle, bus, metro; car-pool |
| Shopping | buy only what you need; repair and reuse |
No poverty, zero hunger, quality education, clean water and sanitation, affordable and clean energy, climate action. (Any four.)
Water shortage - rainwater harvesting and drip irrigation. Energy crisis - renewable energy such as solar and wind. Soil degradation - organic farming and crop rotation. Air pollution - public transport and electric vehicles. Waste - reduce, reuse, recycle.
Ten marks: four one-mark questions (out of six) and three two-mark questions (out of five). One question comes from each unit, and two one-markers from Self-Management.
Clear, Concise, Concrete, Correct, Coherent, Complete, Courteous
Physical (noise), linguistic (language), interpersonal (trust), organisational (rules), cultural
The reply of the receiver. Good feedback is specific, timely, polite and helpful.
Stress is our reaction to a difficult situation.
Specific, Measurable, Achievable, Realistic, Time bound
Organise, Prioritise, Control, Track
| Term | Meaning |
|---|---|
| Operating system | interface between the user and the hardware: Windows, Linux, Android |
| File and folder | a file stores data and has an extension (.txt Notepad, .docx Word, .jpg picture, .mp3 audio); a folder holds files |
| Virus | a program that harms the computer |
| Worm | copies itself and spreads from computer to computer |
| Trojan Horse | looks useful, harms when run; does not copy itself |
| Protection | strong passwords, antivirus, firewall, backup, secure (https) sites |
Patience, positivity, hard work, perseverance, confidence, creativity, open to trial and error
Takes decisions, manages the business, takes risk, brings new ideas, divides the income, creates jobs
Self-employed (own boss) and wage employed (works for a salary). Stages: Enter, Survive, Grow.
| Term | Meaning |
|---|---|
| Sustainable development | meets the needs of the present without harming the ability of future generations to meet theirs |
| SDGs | 17 goals adopted by the United Nations in 2015, to be achieved by 2030 |
| Goals to remember | 1 No Poverty, 2 Zero Hunger, 4 Quality Education, 6 Clean Water and Sanitation, 7 Affordable and Clean Energy, 12 Responsible Consumption and Production, 13 Climate Action |
| Our role | reduce, reuse, recycle; save water and power; plant trees; say no to plastic |
7 marks · five one-markers, one short and one long question
In Class IX you learnt that an AI project is not built in one jump. It moves through a fixed set of stages, and each stage gives the input for the next one. In Class X we use the same cycle again, but now we look at each stage more carefully and we also ask a new question at every stage: is this fair and safe for the people affected? That second question is the subject of the second half of this chapter, the ethical frameworks for AI.
The AI Project Cycle is a step-by-step framework used to plan, build and use an AI solution for a real problem. It has six stages:
It is called a cycle because the stages repeat. If the model fails in Evaluation, we go back to collect more data or to try another model; after Deployment, feedback from users may start a new round of problem scoping.
Problem scoping means understanding the problem so clearly that we can write it in one or two sentences and decide what success will look like. A badly scoped problem wastes all later work: the best model in the world is useless if it solves the wrong problem.
| W | Question we ask | Example: students of a village school drop out after Class VIII |
|---|---|---|
| Who | Who are the stakeholders? Who faces the problem? | students, parents, teachers, the school principal |
| What | What is the problem? What is the evidence? | about 20% students leave school after Class VIII every year |
| Where | In which context, situation or place does it happen? | in families with low income, mostly in the harvest season |
| Why | Why should it be solved? What is the benefit? | education improves jobs and income; achieves SDG 4 |
Our [stakeholders] have a problem that [issue] when/while [context]. An ideal solution would [benefit].
Our students of Classes VI-VIII and their parents have a problem that many students leave school before Class IX when their families need them for farm work in the harvest season. An ideal solution would identify students at risk early so that teachers can counsel families and arrange support, reducing dropouts.
Data acquisition is collecting the data needed to train and test the model. The model learns only from this data, so the quality of the data decides the quality of the model ("garbage in, garbage out").
Collecting data without the permission of the people concerned is unethical even if the project goal is good. Always take consent and protect personal details.
Raw data is often messy: missing values, duplicate rows, spelling differences. In data exploration we clean the data and draw graphs to understand it. Graphs make trends visible that are hidden in long tables.
| Graph | Best for | Example |
|---|---|---|
| Bar graph | comparing categories | dropouts in each village |
| Line graph | change over time | attendance month by month |
| Pie chart | parts of a whole | reasons for leaving school |
| Scatter plot | relation between two variables | attendance vs marks |
| Histogram | how values are spread | number of students in each marks range |
From the graphs we may find, for example, that students whose attendance falls below 60% in August are most likely to drop out. Such insights help us choose features and a model.
A model is a program that takes inputs (features) and gives an output (prediction). There are two broad approaches:
The details of learning-based models - supervised, unsupervised, reinforcement learning and neural networks - are the subject of Unit 2.
The trained model is tested on the testing data, which it has never seen. We compare its predictions with the real answers and calculate measures such as accuracy, precision, recall and F1 score (Unit 3). If the results are poor we go back - more data, better features or a different model.
Deployment means putting the tested model into real use: as a mobile app, a website, a dashboard for teachers or a part of a machine. Deployment also includes training the users, watching the model's performance over time and updating it when the situation changes (for example after a new admission policy).
The dropout model is added to the school's attendance app. Every Monday it shows the class teacher a short list of students at risk with the reason (low attendance, falling marks). The teacher calls the parents. Each term the school checks how many listed students actually stayed in school, and the model is retrained with the new data.
A framework is a set of steps or principles that helps us solve a problem in an organised way. Ethical frameworks are frameworks that help us make sure that the choices we make - while building and using AI - do not cause unintended harm to people.
AI systems take decisions that affect people: who gets a loan, who gets called for an interview, which patient gets extra care. These systems learn from past data, and past data can carry human bias. Without a careful ethical check, an AI can be unfair to some groups, invade privacy or cause harm, even when its designers had good intentions. Ethical frameworks:
Ethical frameworks for AI are of two main types: sector-based and value-based.
These are made for one particular field (sector), because each field has its own risks. The best-known example is bioethics, used in healthcare and medical research. Similar frameworks exist for finance, education and law enforcement.
These are based on basic moral values and can be used in any field. There are three kinds:
| Value-based framework | Main idea | Question it asks | AI example |
|---|---|---|---|
| Rights-based | protect human rights and dignity | Does it respect every person's rights (privacy, equality, freedom)? | a face-recognition system must not track people without consent |
| Utility-based | greatest good for the greatest number | Which choice gives the most benefit and the least harm overall? | a traffic AI chooses signal timings that cut total waiting time for the city |
| Virtue-based | act as a good person would | Would an honest, fair, kind and responsible person do this? | a company refuses to sell a chatbot that manipulates children to buy things |
Rights = respect each person; Utility = count total good and harm; Virtue = the character of the decision-maker.
Bioethics is the study of ethical issues in biology and medicine. It is guided by four principles:
| Principle | Meaning | In an AI for hospitals |
|---|---|---|
| Respect for autonomy | people have the right to make informed decisions about their own body and data | patients are told that AI is used and can refuse |
| Do not harm (non-maleficence) | avoid causing harm | test the AI carefully; a wrong diagnosis can hurt |
| Ensure maximum benefit for all (beneficence) | act for the good of patients | the AI should improve care, not only cut cost |
| Give justice | treat people fairly and share benefits equally | the AI must work equally well for every group |
A large hospital system used an AI algorithm to decide which patients needed extra care (special nurses, more follow-ups). The algorithm was trained to predict future health-care cost, on the idea that sicker patients spend more.
What went wrong? One community of patients had historically spent less money on health care - not because they were healthier, but because of lower income and poorer access to hospitals. The algorithm therefore gave them lower risk scores, and many seriously ill patients from that community were not selected for extra care.
Applying bioethics:
Fix: use direct measures of health (number of active illnesses, test results) instead of cost as the target, test the model separately for every group, and keep doctors in the final decision.
The mistake was in problem scoping and data acquisition (choosing cost as the label), not in the code. Ethical checks must start at the very first stage of the AI project cycle.
| Stage | Ethical question to ask |
|---|---|
| Problem Scoping | Who could be harmed? Are all stakeholders included? |
| Data Acquisition | Is there consent? Is the data representative of all groups? |
| Data Exploration | Are some groups missing or under-represented? |
| Modelling | Is the model explainable? Does it use unfair features (caste, religion, gender)? |
| Evaluation | Is accuracy equal for every group? |
| Deployment | Can people appeal a decision? Who is accountable for mistakes? |
Sector-based frameworks are designed for a specific field, e.g. bioethics for healthcare. Value-based frameworks are based on general moral values and can be used in any field; they are rights-based, utility-based and virtue-based.
Respect for autonomy, do not harm, ensure maximum benefit for all, give justice. Respect for autonomy: patients decide about their own treatment and data after being informed. Give justice: benefits and burdens of health care are shared fairly among all groups; an AI must not favour one group.
Utility-based framework - it chooses the action with the greatest good for the greatest number.
| Stage | What is done | Remember |
|---|---|---|
| Problem Scoping | understand the problem and set the goal | 4Ws canvas: Who, What, Where, Why |
| Data Acquisition | collect reliable, relevant data | surveys, sensors, cameras, web, open portals |
| Data Exploration | arrange the data and draw graphs | to see trends and patterns |
| Modelling | choose and train the AI model | rule-based or learning-based |
| Evaluation | test the model on new data | accuracy, precision, recall, F1 |
| Deployment | put the model to real use | app, website, device |
Numbers and tables.
Price prediction, weather forecastImages and videos.
Face unlock, self-driving carsText and speech.
Voice assistants, translationAn ethical framework is a set of values and principles that helps us decide what is right while building and using AI, so that the system is fair and does not harm people.
In daily life: School rules for the science lab are sector-based. The values of honesty and fairness, which apply everywhere, are value-based.
| Framework | Asks | Example |
|---|---|---|
| Rights-based | Does the action protect the rights and dignity of every person? | an AI must not use personal data without consent |
| Utility-based | Does it give the greatest good to the greatest number? | a traffic AI that saves the most travel time for the city |
| Virtue-based | Would an honest, kind and fair person do it? | a company openly reports the mistakes of its AI |
| Bioethics (sector-based) | health care: respect for autonomy, do not harm, ensure maximum benefit for all, give justice | a diagnosis AI tested on patients of all groups |
11 marks · the heaviest AI unit
Artificial Intelligence (AI) is the broad idea of machines that act intelligently: they sense, learn, reason and decide. Machine Learning (ML) is a part of AI in which machines learn from data instead of following only hand-written rules. Deep Learning (DL) is a part of ML that uses artificial neural networks with many layers and very large data.
| AI | ML | DL | |
|---|---|---|---|
| Scope | widest | part of AI | part of ML |
| How it works | any technique that shows intelligent behaviour, including rules | learns patterns from data | learns features by itself using many-layer neural networks |
| Data needed | may need none (rule-based) | moderate | very large |
| Example | a chess program with rules | a spam filter | face recognition, voice assistants |
In the rule-based approach the developer studies the problem and writes rules: *if temperature > 38 and cough = yes then "see a doctor"*. The machine follows the rules but cannot learn new ones. In the learning-based approach the developer gives examples (data) and the machine discovers the rules.
| Point | Rule-based | Learning-based |
|---|---|---|
| Who makes the rules? | the programmer | the machine, from data |
| Learns from new data? | no | yes |
| Handles new situations | poorly | better, if trained on similar data |
| Explainability | easy to explain | can be hard to explain |
| Example | a fixed-reply chatbot, a grading rule | spam filter, movie recommendations |
Machine learning models are grouped by the kind of data and feedback they learn from.
The model learns from labelled data - like a student learning with an answer key. It learns the mapping from features to label and then predicts labels for new data.
(i) Predict whether it will rain tomorrow - classification (yes/no). (ii) Predict how many millimetres of rain will fall - regression (a number). (iii) Identify the breed of a dog in a photo - classification. (iv) Predict the price of a used car - regression.
The model gets unlabelled data and finds hidden structure by itself - like sorting a box of mixed buttons without being told the groups.
An agent learns by trial and error in an environment. Each action gives a reward (good) or a penalty (bad), and the agent learns to choose actions that give the most reward over time - like training a pet with treats.
Examples: game-playing AI (AlphaGo), robots learning to walk, self-driving car simulations, ad placement.
| Type | Data | Goal | Example |
|---|---|---|---|
| Supervised | labelled | predict a label | spam detection, price prediction |
| Unsupervised | unlabelled | find groups or patterns | customer segmentation |
| Reinforcement | rewards / penalties | learn the best actions | game AI, robots |
Clustering is NOT classification. Classification needs labels given in advance; clustering creates its own groups from unlabelled data.
Artificial neural networks (ANNs) are inspired by the human brain, where billions of neurons pass signals to one another. An artificial neuron (node) takes several inputs, multiplies each by a weight, adds a bias, and passes the result through an activation function to produce an output.
| Layer | Role |
|---|---|
| Input layer | receives the raw data (for an image, one node per pixel); does no processing |
| Hidden layer(s) | do the processing; each node combines inputs with weights and passes the result on; "deep" networks have many hidden layers |
| Output layer | gives the final result, e.g. one node for "cat" and one for "dog" |
Inputs: hours studied x1 = 4, attendance x2 = 0.9. Weights w1 = 0.5, w2 = 2, bias = -1.5. Weighted sum = 0.5 x 4 + 2 x 0.9 - 1.5 = 2 + 1.8 - 1.5 = 2.3. With a step activation "output 1 if sum > 0", the neuron outputs 1 (pass).
| Problem | Type | Why |
|---|---|---|
| Will a customer leave the bank? | supervised - classification | past customers are labelled left / stayed |
| How many units will we sell next month? | supervised - regression | output is a number |
| Group 10,000 shoppers for offers | unsupervised - clustering | no labels given |
| Teach a robot arm to pick objects | reinforcement | it learns from success and failure |
| Recognise handwritten digits | supervised - classification (often a neural network) | labelled images of digits |
The handbook encourages trying models without coding:
Make two classes, "Thumbs up" and "Thumbs down". Record about 50 webcam images of each (the labelled training data). Click Train. Show your hand to the camera: the model predicts the class with a confidence percentage. If you show a new gesture it was never trained on, it will still pick one of the two classes - showing that a model only knows what it was trained on.
Supervised learning trains on labelled data to predict labels for new data, e.g. classifying e-mails as spam. Unsupervised learning works on unlabelled data and finds hidden groups, e.g. grouping customers by buying behaviour.
A neural network has an input layer, one or more hidden layers and an output layer. The input layer receives the data, one node per feature (e.g. each pixel). Hidden layers process the data: each node multiplies inputs by weights, adds a bias and applies an activation function; many hidden layers make the network "deep". The output layer gives the final prediction, e.g. the probability of each class. The network learns by adjusting weights to reduce the error between predicted and actual outputs.
Reinforcement learning.
In daily life: Fruits, mangoes and Alphonso mangoes: every Alphonso is a mango and every mango is a fruit.
| Term | Meaning | Example: will a student pass? |
|---|---|---|
| Features | the input columns | attendance, study hours |
| Label | the output to be predicted | pass or fail |
| Training data | used to teach the model | 80% of the records |
| Testing data | kept aside to check the model | 20% of the records |
The developer writes the rules. The machine only follows them.
Fixed-reply chatbot. Does not learn from new data.The machine finds the patterns from data on its own.
Spam filter. Improves with more data.| Type | Data | Sub-categories | Example |
|---|---|---|---|
| Supervised | labelled | Classification (category), Regression (number) | spam or not spam; price of a house |
| Unsupervised | unlabelled | Clustering (groups), Association (items together), anomaly detection | groups of customers; bread and butter; fraud |
| Reinforcement | rewards and penalties | trial and error | game-playing AI, robot learning to walk |
In daily life: A relay race: the first runner takes the baton (input), the middle runners carry it (hidden), the last runner finishes (output).
10 marks · six one-markers, one short question and the confusion matrix numerical
Evaluation is the process of checking how well a trained AI model performs on data it has not seen before. It is the fifth stage of the AI project cycle and decides whether the model is ready for deployment.
Why is evaluation necessary?
Never evaluate a model on its own training data. It is like giving students the exact questions they practised: the marks will look high but will not show real understanding.
Before training, the available labelled data is divided into two parts:
| Part | Used for | Typical share |
|---|---|---|
| Training set | teaching the model the patterns | 70% to 80% |
| Testing set | checking the trained model on unseen data | 20% to 30% |
A dataset has 1,000 labelled X-ray images. With an 80:20 split, 800 images are used to train and 200 are kept aside to test.
Accuracy alone can be misleading. Suppose only 2 out of 100 forest areas actually catch fire. A lazy model that always predicts "no fire" is correct 98 times - 98% accuracy - but it never warns about a single real fire! We therefore need a more detailed picture: the confusion matrix.
For a classification problem with two classes (Positive and Negative), every prediction falls into one of four boxes:
| Reality: Yes (Positive) | Reality: No (Negative) | |
|---|---|---|
| Prediction: Yes | True Positive (TP) - correctly predicted Yes | False Positive (FP) - predicted Yes, actually No (false alarm) |
| Prediction: No | False Negative (FN) - predicted No, actually Yes (missed case) | True Negative (TN) - correctly predicted No |
The words are easy to decode: True / False says whether the prediction was right; Positive / Negative says what the model predicted.
The model predicts "fire" and there really is a fire: TP. It predicts "fire" but there is none: FP (fire brigade rushes for nothing). It predicts "no fire" but there is a fire: FN (dangerous - the fire spreads). It predicts "no fire" and there is none: TN.
All four measures below come from the confusion matrix.
The choice depends on which mistake is more costly.
| Situation | Costlier mistake | Metric to focus on | Reason |
|---|---|---|---|
| Forest fire / viral outbreak detection | FN (missing a real fire or outbreak) | Recall | a missed case can cause huge damage |
| Cancer screening test | FN | Recall | a missed patient goes untreated |
| Spam filter | FP (a real e-mail goes to spam) | Precision | losing an important mail is worse than seeing one spam |
| Mining - predicting where to dig for gold | FP (digging at the wrong place) | Precision | each wrong dig costs crores |
| Balanced situations | both | F1 score | balances precision and recall |
if missing a positive is dangerous - recall; if a false alarm is costly - precision.
A model tested on 200 cases gives TP = 60, FP = 20, FN = 10, TN = 110. Find accuracy, precision, recall and F1 score.
An AI checks 100 patients for a disease. 30 really have the disease; the AI correctly finds 24 of them. Of the 70 healthy people, the AI wrongly says 6 have the disease.
Out of 1,000 transactions only 10 are fraud. Model A always says "not fraud": TP = 0, FN = 10, FP = 0, TN = 990. Accuracy = 990 / 1000 = 99%, but Recall = 0 / 10 = 0 - it catches no fraud at all. Model B gives TP = 8, FN = 2, FP = 40, TN = 950: accuracy 95.8% (lower) but recall 0.8 - far more useful.
A confusion matrix is a 2 x 2 table that compares a model's predictions with the actual values. Its cells are TP, FP, FN and TN. All evaluation metrics - accuracy, precision, recall and F1 - are calculated from it.
Recall, because a False Negative (an actual outbreak predicted as no outbreak) means no preparation is made and the virus spreads; missing real cases is far more costly than a false alarm.
Model evaluation means checking how well a trained model works, before it is deployed. The model is always tested on data it has not seen in training.
Confusion matrix
| Reality: Yes | Reality: No | |
|---|---|---|
| Predicted: Yes | 40TP | 10FP |
| Predicted: No | 5FN | 45TN |
| Metric | Formula | With TP = 40, FP = 10, FN = 5, TN = 45 |
|---|---|---|
| Accuracy | (TP + TN) / Total | (40 + 45) / 100 = 85 / 100 = 0.85 (85%) |
| Precision | TP / (TP + FP) | 40 / (40 + 10) = 40 / 50 = 0.80 (80%) |
| Recall | TP / (TP + FN) | 40 / (40 + 5) = 40 / 45 = 0.89 (88.9%) |
| F1 score | 2 x Precision x Recall / (Precision + Recall) | 2 x 0.80 x 0.89 / (0.80 + 0.89) = 1.4222 / 1.69 = 0.84 |
When a False Positive is costly.
Spam filter: a genuine mail must not go to spam.When a False Negative is costly.
Disease test, fire alarm: no real case should be missed.Actual value = 400 Predicted value = 380 Error = |400 - 380| = 20 Error rate = 20 / 400 = 0.05 = 5% Accuracy = 100% - 5% = 95%
4 marks · four one-markers and one short question
Computer Vision (CV) is the domain of AI that enables machines to see, understand and interpret images and videos - and take decisions from them, just as humans do with their eyes and brain. A camera only captures light; computer vision gives meaning to what is captured: "this is a cat", "the traffic light is red", "this X-ray shows a fracture".
| Image processing | Computer vision | |
|---|---|---|
| Goal | improve or change an image | understand what is in the image |
| Output | another image | information or a decision |
| Example | brightening, cropping, removing noise, filters | recognising faces, reading number plates |
| Application | What CV does |
|---|---|
| Facial recognition | unlocks phones, marks attendance, tags friends in photos |
| Face filters | detects eyes, nose and mouth to place filters (Instagram, Snapchat) |
| Google Lens / search by image | identifies objects, plants, products and text from a photo |
| Self-driving cars | detects lanes, signals, pedestrians and other vehicles |
| Medical imaging | finds tumours, fractures and eye disease in scans |
| Google Translate camera | reads text in a photo and translates it on screen |
| Retail | tracks stock on shelves, automatic billing counters |
| Agriculture | detects crop disease and weeds from drone photos |
| Security | CCTV analysis, number-plate recognition at toll plazas |
one label for the image
label + one box
many objects, many boxes
exact outline of each object
CV tasks are grouped by how many objects are involved and how precisely they are located.
| Task | Objects | Output | Example |
|---|---|---|---|
| Image classification | single | one label for the whole image | "this photo is a dog" |
| Classification + localisation | single | label + a bounding box around the object | "dog, here" |
| Object detection | multiple | label + box for each object | all cars and people in a street photo |
| Instance segmentation | multiple | exact outline (pixel mask) of each object | separating each person in a crowd, pixel by pixel |
classification < classification + localisation < object detection < instance segmentation.
A digital image is a grid of tiny squares called pixels (picture elements). Each pixel stores a number (or three numbers) for its colour.
An image is 640 pixels wide and 480 pixels high. Number of pixels = 640 x 480 = 307,200. If it is grayscale, it needs 307,200 values; if it is a colour (RGB) image, it needs 3 x 307,200 = 921,600 values.
| RGB value | Colour |
|---|---|
| (255, 0, 0) | red |
| (0, 255, 0) | green |
| (0, 0, 255) | blue |
| (255, 255, 0) | yellow (red + green) |
| (0, 255, 255) | cyan (green + blue) |
| (255, 0, 255) | magenta (red + blue) |
| (128, 128, 128) | grey |
Number of possible colours in 24-bit RGB = 256 x 256 x 256 = 16,777,216 (about 16.7 million).
A feature is a piece of information in an image that helps to identify objects. Good features are distinctive and easy to find again in another image.
When solving a jigsaw, pieces of plain sky are hard to place, edge pieces are easier, and corner pieces are the easiest. In the same way, for a computer, corners are the best features, then edges, while flat regions give little information.
Convolution is the operation used to find features. A small matrix called a kernel (filter), e.g. 3 x 3, is slid over the image. At each position the overlapping pixel values are multiplied by the kernel values and added; the sum becomes one pixel of the output (the feature map). Different kernels detect different features: one finds vertical edges, another horizontal edges, another sharpens or blurs.
Image patch: [[2, 2, 2], [0, 0, 0], [5, 5, 5]], kernel: [[1, 1, 1], [0, 0, 0], [-1, -1, -1]] (a horizontal-edge detector). Multiply element by element and add: (2+2+2) + 0 + (-5-5-5) = 6 - 15 = -9. A large value (positive or negative) means a strong edge here.
A CNN is a neural network designed for images. Its layers are:
| Layer | What it does |
|---|---|
| Convolution layer | applies many kernels to find features such as edges and textures, creating feature maps |
| ReLU (Rectified Linear Unit) layer | replaces negative values with 0, keeping only useful positive signals and adding non-linearity |
| Pooling layer | shrinks the feature maps (e.g. max pooling keeps the largest value in each 2 x 2 block), making the network faster and less sensitive to small shifts |
| Fully connected layer | combines all the features and gives the final class probabilities, e.g. cat 0.92, dog 0.08 |
Early layers find simple features (edges), middle layers find shapes (eyes, wheels), and the last layers recognise whole objects (faces, cars).
Object detection finds multiple objects and draws a bounding box with a label around each. Instance segmentation goes further and marks the exact pixel outline of every object, separating even objects of the same class.
0 to 255; 0 is black (no intensity) and 255 is white (full intensity).
Computer Vision is the domain of AI that lets a machine see, process and understand images and videos. Uses: face recognition, self-driving cars, medical imaging, search by image, number-plate reading.
Each small square is a pixel (picture element). In an 8-bit grayscale image each pixel has a value from 0 (black) to 255 (white).
A colour (RGB) pixel stores three such values: red, green and blue. (255, 0, 0) is red; (255, 255, 0) is yellow; (255, 255, 255) is white.
| Term | Meaning | Numerical |
|---|---|---|
| Resolution | number of pixels, width x height | 640 x 480 = 3,07,200 pixels |
| Grayscale image | 1 channel, 1 byte per pixel | 640 x 480 image needs 3,07,200 bytes |
| RGB image | 3 channels, 3 bytes per pixel | 3,07,200 x 3 = 9,21,600 bytes |
| 8-bit pixel | 2 to the power 8 shades | 256 values: 0 to 255 |
| Term | Meaning |
|---|---|
| Image features | corners and edges are good features; flat patches are not |
| Convolution | a small matrix (kernel) slides over the image and brings out features such as edges |
| CNN layers | Convolution layer, ReLU layer, Pooling layer (makes the feature map smaller), Fully connected layer (gives the class) |
| No-code tools | Teachable Machine, Lobe, Orange Data Mining |
8 marks · five one-markers, one short question and the Bag of Words numerical
Natural Language Processing (NLP) is the domain of AI that helps computers understand, interpret and generate human language - spoken or written. Human ("natural") languages such as Hindi or English are full of ambiguity and variety, while computers understand only exact, structured instructions. NLP bridges this gap.
| Application | Example |
|---|---|
| Automatic summarisation | news apps give a short summary of a long article |
| Sentiment analysis | companies find whether reviews and tweets are positive, negative or neutral |
| Text classification | spam filtering; sorting complaints by department |
| Virtual assistants | Siri, Alexa, Google Assistant understand spoken commands |
| Machine translation | Google Translate, Bhashini for Indian languages |
| Chatbots | customer-support bots on bank and railway sites |
| Autocomplete and grammar check | Gmail Smart Compose, Grammarly |
A chatbot is a program that talks with users in natural language.
| Script bot | Smart bot | |
|---|---|---|
| Working | follows a fixed script and rules | uses AI and NLP; learns from data |
| Flexibility | limited; fails on unexpected questions | understands varied questions |
| Building | easy, little or no coding | needs data, training and coding |
| Memory / learning | none | improves with use |
| Example | a bank's menu bot ("press 1 for balance") | ChatGPT, Google Assistant, Alexa |
| Human language | Computer language |
|---|---|
| ambiguous: one word can have many meanings | exact: each instruction has one meaning |
| grammar is flexible; small mistakes are understood | any syntax error stops the program |
| meaning depends on context, tone and culture | context-free |
Two ideas explain why understanding language is hard:
Before a computer can work with text, the text is cleaned and simplified - text normalisation. The collection of all the documents used is called the corpus.
"Raj and Vijay are best friends! They play together." -> segmentation: 2 sentences -> tokens: Raj, and, Vijay, are, best, friends, !, They, play, together, . -> remove stop words and punctuation: Raj, Vijay, best, friends, play, together -> lower case: raj, vijay, best, friends, play, together -> stemming/lemmatisation: friends -> friend.
| Word | Stem | Lemma |
|---|---|---|
| studies | studi | study |
| studying | study | study |
| caring | car | care |
| healed | heal | heal |
| better | better | good |
Bag of Words is an NLP model that represents text by the frequency of each word, ignoring grammar and word order - as if all the words were thrown into a bag and counted.
Steps of the BoW algorithm:
Corpus after normalisation - Document 1: "aman and anil are stressed"; Document 2: "aman went to a therapist"; Document 3: "anil went to download a health chatbot". Dictionary: aman, and, anil, are, stressed, went, to, a, therapist, download, health, chatbot.
| Document | aman | and | anil | are | stressed | went | to | a | therapist | download | health | chatbot |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| D1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| D2 | 1 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 | 0 | 0 | 0 |
| D3 | 0 | 0 | 1 | 0 | 0 | 1 | 1 | 1 | 0 | 1 | 1 | 1 |
Bag of Words treats common words and rare words alike. TF-IDF gives a word more importance if it occurs often in one document but rarely in the others - such words tell us what that document is about.
Applications of TF-IDF: document classification, topic modelling, information retrieval (search engines rank pages), keyword extraction and stop-word filtering.
Tools such as Orange Data Mining (text widgets: corpus, preprocess text, bag of words, word cloud, sentiment analysis) let students try NLP without coding. In a sentiment-analysis activity, product reviews are loaded, normalised, converted to a bag of words and labelled positive or negative.
Stemming removes affixes to get a stem, which may not be a meaningful word (caring -> car); it is fast. Lemmatisation gives the dictionary base word (lemma), which is always meaningful (caring -> care); it is slower but more accurate.
IDF = log(N / DF) = log(N / N) = log 1 = 0, because such a word does not help to distinguish one document from another.
Natural Language Processing is the domain of AI that lets a machine understand and produce human language. Uses: voice assistants, language translation, sentiment analysis, text classification, keyword extraction, automatic captions.
Works on a fixed script of rules. Easy to make, limited use.
Menu-based customer care botUses AI, learns from data, handles new questions.
Alexa, Google Assistant| Level | Meaning | Example |
|---|---|---|
| Syntax | grammar: the arrangement of words | "Chickens feed extravagantly while the moon drinks tea" has correct syntax but no meaning |
| Semantics | meaning of the words and sentence | "bank" of a river and money in the "bank" |
| Step | Work | Example |
|---|---|---|
| Tokenisation | breaks a sentence into tokens: words, numbers and special characters | "I love AI!" gives 4 tokens: I | love | AI | ! |
| Stop words | very frequent words with little meaning are removed | a, an, the, is, are, to, and |
| Stemming | cuts the affixes; the result may not be a word | studies gives studi |
| Lemmatisation | gives the meaningful base word (lemma); slower | studies gives study; better gives good |
Bag of Words counts how many times each word of the vocabulary occurs in each document. The order of the words is not kept.
Step 1. Text normalisation: lower case, punctuation removed. Step 2. Dictionary (vocabulary) of 9 words: aman, likes, to, play, cricket, riya, chess, and, together. Step 3 and 4. Document vectors:
| aman | likes | to | play | cricket | riya | chess | and | together | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 1 | 1 | 1 | 0 | 1 | 1 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 1 | 1 |
| Word | aman | likes | to | play | cricket | riya | chess | and | together |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 2 | 3 | 1 | 2 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/2 | 3/2 | 3/2 | 3/3 | 3/1 | 3/2 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.176 | 0.176 | 0.176 | 0.000 | 0.477 | 0.176 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log( IDF(W) ) IDF = total documents / document frequency 'cricket' in Document 1: 1 x log(3/1) = 1 x 0.477 = 0.477 (rare word, high value) 'aman' in Document 1: 1 x log(3/2) = 1 x 0.176 = 0.176 'play' in Document 1: 1 x log(3/3) = 1 x 0 = 0 (in every document, no value)
684 questions in 15 types, including 187 numericals with full working
Tap a type to open it. In the one-mark types, tap an option to check yourself. The answers of all the numericals were worked out by a program.
In the communication cycle, changing the idea into words, pictures or gestures is called
(b) encoding Encoding turns the sender's idea into a message (words, pictures, gestures); decoding is the receiver understanding it.
The receiver's reply that tells the sender the message is understood is called
(b) feedback Feedback completes the communication cycle by telling the sender how the message was received.
Which of these is NOT a part of the 7 Cs of communication?
(a) Costly The 7 Cs are Clear, Concise, Concrete, Correct, Coherent, Complete and Courteous; 'Costly' is not one of them.
A road sign showing a zebra crossing is an example of
(d) visual communication It sends the message through a picture, so it is visual communication.
A smile, a nod and a handshake are examples of
(b) non-verbal communication They communicate without words, using body language, so they are non-verbal.
Ravi cannot understand his new classmate because they speak different languages. This is a
(c) linguistic barrier A difference of language is a linguistic (language) barrier.
Loud traffic noise outside the classroom disturbs a lesson. This is a
(a) physical barrier Noise in the surroundings is a physical (environmental) barrier.
"Well done! Your chart is neat and the headings are clear." This is an example of
(a) specific (descriptive) feedback It praises and also says exactly what was good, so it is specific (descriptive) feedback.
Which type of sentence gives an order or makes a request?
(a) Imperative Imperative sentences give orders or requests, e.g. "Close the door."
In the sentence "Divya reads a book", the word 'reads' is the
(d) verb 'reads' shows the action, so it is the verb.
E-mail, SMS and letters are forms of
(c) written verbal communication They use written words, so they are written verbal communication.
Which of the 7 Cs asks us to use only as many words as needed?
(c) Concise Concise means short and to the point.
Riya practises hard only because the best performer will get a trophy. Her motivation is
(b) external motivation A prize, praise or reward from others is external motivation.
Our emotional, mental and physical reaction to a difficult situation is called
(c) stress Stress is the body's and mind's reaction to pressure.
In SMART goals, the letter M stands for
(d) Measurable M = Measurable: progress can be counted or checked.
In SMART goals, the letter T stands for
(b) Time bound T = Time-bound: the goal has a deadline.
"I will improve" is a weak goal. Which is a SMART goal?
(c) I will score 80% in the AI pre-board in January by studying one hour daily It is Specific (AI pre-board), Measurable (80%), Achievable, Relevant and Time-bound (January, one hour daily).
Aman learns the guitar because he enjoys music. This is
(c) internal motivation He is driven by his own enjoyment, so it is internal (intrinsic) motivation.
Knowing your own strengths, weaknesses, likes and dislikes is called
(d) self-awareness Self-awareness means knowing yourself.
Which of these is a stress management technique?
(d) Regular physical exercise Exercise releases tension and improves mood; it is a healthy way to manage stress.
The ability to control your behaviour, emotions and thoughts in the long-term interest is
(c) self-regulation Self-regulation (self-control) is managing your own behaviour and emotions.
Which of these is the first step of time management?
(c) Organise: plan the day's activities First organise (plan) the activities; then prioritise, control and track.
Doing the most important task first is called
(c) prioritising Prioritising means doing tasks in order of importance.
Which quality is needed to work independently?
(c) Self-motivation Self-motivation keeps a person working without being pushed by others.
Which of these is a sign of stress?
(d) Headache and sleeplessness Headache and sleeplessness are physical signs of stress.
A student feels nervous before the board exam. Which is the best way to manage it?
(b) Make a time table and take short breaks A time table removes confusion and short breaks keep the mind fresh.
In SMART goals, 'A' stands for
(c) Achievable A = Achievable: the goal can really be reached.
Which of these is NOT a step of time management?
(d) Postpone Postponing is procrastination, the opposite of good time management.
Good stress that pushes us to perform better, such as before a match, is called
(c) eustress (positive stress) Eustress is positive stress that motivates us.
Which habit builds self-confidence?
(b) Getting feedback and working on it Working on feedback improves skills, which builds confidence.
Finding where your time was spent during the day is called
(b) tracking Tracking records how time was used, so it can be improved.
Yoga and meditation mainly help in
(c) reducing stress They calm the mind and body, which reduces stress.
A person who can manage himself or herself well
(b) sets goals and works to achieve them Self-management means setting goals and working steadily towards them.
Which is the default file extension of a file made in Notepad?
(d) .txt Notepad saves plain text files with the extension .txt.
The software that works as an interface between the user and the computer hardware is the
(a) operating system The operating system manages the hardware and lets the user and programs use it.
Which of these is an operating system for mobile phones?
(b) Android Android (by Google) runs on most mobile phones.
A program that looks useful but harms the computer when it is run is a
(a) Trojan Horse A Trojan Horse hides harmful code inside a program that looks useful.
A malicious program that copies itself from one computer to another without human help is a
(a) worm A worm spreads by itself through networks.
Which of these protects a computer from viruses?
(d) Antivirus software Antivirus software detects and removes viruses.
A copy of important files kept at another place is called a
(c) backup A backup lets you recover files if the original is lost or damaged.
Which is the strongest password?
(d) Tr@in#2027kv It mixes capital and small letters, digits and symbols and is long.
A website address that begins with https shows that the
(b) data sent to the site is encrypted The S means secure: the data sent is encrypted.
Which key combination copies the selected file or text?
(d) Ctrl + C Ctrl + C copies, Ctrl + X cuts and Ctrl + V pastes.
Stealing someone's personal details to cheat others in that person's name is
(a) identity theft Using another person's identity to cheat is identity theft.
Which of these is a good way to care for a computer?
(c) Shut it down properly and keep it dust-free Proper shutdown and a clean, dust-free machine keep it working longer.
A person who starts a business by taking a risk to earn profit and solve a problem is an
(b) entrepreneur An entrepreneur takes risk to start a business that solves a problem.
A person who works for a salary under an employer is
(b) wage employed Wage employment means working for someone else for a salary.
Which of these is a quality of a successful entrepreneur?
(d) Perseverance: never giving up Perseverance means not giving up after failures.
"Entrepreneurs are born, not made." This statement is a
(d) myth It is a myth: entrepreneurial skills can be learnt and developed.
Which of these is a function of an entrepreneur?
(a) Taking decisions and managing the business An entrepreneur takes decisions and manages the business.
Finding a new method or product to solve a problem is called
(a) innovation Innovation is creating new ideas, methods or products.
Which is the correct order of the stages of an entrepreneur's career?
(d) Enter, Survive, Grow An entrepreneur first enters (starts), then survives the early years, then grows the business.
Which of these is NOT a role of an entrepreneur?
(b) Refusing to take any risk Entrepreneurs must take calculated risks; refusing all risk is not their role.
Meena opens a tiffin service that employs five women of her village. She helps society by
(d) creating jobs Her business gives jobs to others, which helps society.
"A business needs a lot of money to start." This statement is a
(b) myth about entrepreneurship It is a myth: many businesses start small with little money.
The ability to keep working even after failing is called
(b) perseverance Perseverance is continuing in spite of difficulties.
A person who runs his or her own small shop is
(b) self-employed Working for oneself is self-employment.
A company makes spoons that can be eaten and that decay naturally. Which Sustainable Development Goal does this support most directly?
(d) SDG 12: Responsible Consumption and Production Edible, biodegradable spoons cut plastic waste: responsible consumption and production (SDG 12).
How many Sustainable Development Goals were adopted by the United Nations in 2015?
(a) 17 The UN adopted 17 SDGs in 2015.
The Sustainable Development Goals are to be achieved by the year
(d) 2030 The SDGs are to be met by 2030 (Agenda 2030).
Development that meets the needs of the present without harming the ability of future generations to meet their needs is
(c) sustainable development This is the definition of sustainable development.
Which SDG is about Quality Education?
(b) SDG 4 SDG 4 is Quality Education.
Using solar panels on the school roof supports
(c) SDG 7: Affordable and Clean Energy Solar power is clean, affordable energy: SDG 7.
Carrying a cloth bag to the market follows which of the 3 Rs?
(a) Reduce (and reuse) It reduces the use of plastic bags (and the cloth bag is reused).
Which SDG is called Climate Action?
(c) SDG 13 SDG 13 is Climate Action.
Which of these is a sustainable practice?
(d) Rainwater harvesting Rainwater harvesting saves water for the future.
Which SDG is about Clean Water and Sanitation?
(d) SDG 6 SDG 6 is Clean Water and Sanitation.
Which of these is a problem that sustainable development tries to solve?
(a) Fast use of natural resources Natural resources are being used up too fast; sustainable development tries to slow this down.
Making new paper from old newspapers is an example of
(b) recycling Turning waste into new material is recycling.
What is the main purpose of the Modelling stage of the AI Project Cycle?
(c) To choose or build an AI model and train it on the data In Modelling a suitable AI model is chosen or built and trained on the data.
Which activities belong to the Data Exploration stage?
(d) Arranging the data and drawing graphs to see patterns and trends Data Exploration arranges the data and uses graphs to find patterns and trends.
The ethical principle 'do not harm' means that an AI system should
(c) not cause injury or damage to people Non-maleficence ('do no harm') means the AI must not hurt people or cause damage.
Which action follows an ethical framework for AI?
(a) Checking the training data for bias before using the model Checking for bias makes the AI fair, which an ethical framework requires.
Which is the first stage of the AI Project Cycle?
(d) Problem Scoping Problem Scoping comes first: understand and define the problem.
In the 4Ws problem canvas, which W asks about the people facing the problem?
(c) Who Who = the stakeholders affected by the problem.
Collecting data from surveys, sensors, cameras and websites is the stage of
(d) Data Acquisition Data Acquisition means collecting data.
Which stage comes just after Modelling?
(c) Evaluation After Modelling the model is evaluated on test data.
Making the tested model available to real users is called
(c) Deployment Deployment puts the model into real use.
Which is the correct order?
(b) Problem Scoping, Data Acquisition, Data Exploration, Modelling, Evaluation, Deployment Problem Scoping → Data Acquisition → Data Exploration → Modelling → Evaluation → Deployment.
Which AI domain works with numbers and tables to find patterns?
(b) Statistical Data (Data Science) Statistical Data (Data Science) works with numerical and tabular data.
A face-unlock feature of a phone belongs to which AI domain?
(d) Computer Vision It recognises faces in camera images, so it is Computer Vision.
A voice assistant that understands spoken questions belongs to which AI domain?
(c) Natural Language Processing Understanding human language is NLP.
A price-comparison website that predicts the best time to buy uses mainly
(a) Statistical Data It studies price data and trends: Statistical Data.
A set of values and principles that helps us decide what is right and wrong while building AI is an
(a) ethical framework An ethical framework guides right decisions while building AI.
Ethical frameworks are broadly of two types:
(c) sector-based and value-based They are sector-based (e.g. bioethics) and value-based (rights, utility, virtue).
Bioethics is an example of a
(a) sector-based framework Bioethics is made for one sector (health care), so it is sector-based.
Bioethics is the ethical framework used in
(d) health care and life sciences Bioethics guides health care and life sciences.
Which of these is NOT a type of value-based framework?
(a) Sector-based Value-based frameworks are rights-based, utility-based and virtue-based; sector-based is the other main type.
A framework that chooses the action giving the greatest good to the largest number of people is
(d) utility-based Utility-based ethics aims at the greatest good for the greatest number.
A framework that puts human rights and dignity first is
(d) rights-based Rights-based ethics protects human rights and dignity.
A framework that looks at the character and intention of the person who acts is
(b) virtue-based Virtue-based ethics looks at good character and intentions.
Respecting a patient's right to take his or her own decision is the principle of
(b) respect for autonomy Autonomy means the patient decides for himself or herself.
Giving every group of patients a fair share of benefits and care is the principle of
(d) justice Justice means fairness for all groups.
A hospital AI was trained mostly on data of one group and gives poor results for others. The main ethical problem is
(a) bias in the training data The data did not represent all groups, so the model is biased.
Why are ethical frameworks needed for AI?
(a) To avoid unintended harm and make fair, responsible decisions They help avoid unintended harm and make AI decisions fair and responsible.
The 4Ws problem canvas is used in which stage?
(d) Problem Scoping The 4Ws (Who, What, Where, Why) are filled in Problem Scoping.
Which of these is a reliable source for data acquisition?
(b) An open government data portal Open government data portals give authentic, checked data.
The 4Ws of the problem canvas are
(b) Who, What, Where, Why Who, What, Where and Why.
Testing the model on new data to see how well it works is the stage of
(b) Evaluation Evaluation checks the model on unseen test data.
Checking whether a loan-approval AI treats all applicants equally is mainly a question of
(d) fairness Treating everyone equally is fairness.
Which of the following is one of the four principles of bioethics?
(a) Ensure maximum benefit for all The four principles are respect for autonomy, do not harm, ensure maximum benefit for all, and justice.
A problem statement template is written at the end of
(c) Problem Scoping After the 4Ws, the problem statement template sums up the problem at the end of Problem Scoping.
Data shown as a bar graph or scatter plot during the project belongs to
(d) Data Exploration Graphs that show patterns are made in Data Exploration.
A game-playing AI improves by playing against itself; it gets points for good moves and loses points for bad moves. This is
(d) reinforcement learning Learning from rewards and penalties is reinforcement learning.
A music app groups songs that sound alike, with no tags given. This is an example of
(c) Unsupervised learning No labels are given, so grouping similar songs is unsupervised learning.
Learning from unlabelled data to find hidden groups is called
(c) unsupervised learning Unsupervised learning finds patterns in unlabelled data.
In a neural network, the processing is done mainly in the
(b) hidden layers The hidden layers transform the inputs and do the main processing.
A neuron has the inputs x1 = 5 with weight w1 = -2, x2 = 1 with weight w2 = -1, x3 = 2 with weight w3 = 4. The bias is -2. What is the weighted sum (with bias)?
(d) -5 Weighted sum = (5 x -2) + (1 x -1) + (2 x 4) + (-2) = -5.
A neuron has the inputs x1 = 0 with weight w1 = 4, x2 = 4 with weight w2 = 1, x3 = 2 with weight w3 = -1 and bias -1. It gives the output 1 if the weighted sum is 2 or more, otherwise 0. What is the output?
(a) 0 Weighted sum = (0 x 4) + (4 x 1) + (2 x -1) + (-1) = 1. It is less than the threshold 2.
The data kept aside to check a trained model is called
(b) testing data Testing data is not used in training; it checks the model.
Clustering and association are types of
(d) unsupervised learning Both work on unlabelled data, so they are unsupervised learning.
Predicting the price of a house (a number) is a problem of
(d) regression Predicting a continuous number is regression.
A neuron has the inputs x1 = 0 with weight w1 = 4, x2 = 4 with weight w2 = 2. The bias is -1. What is the weighted sum (with bias)?
(d) 7 Weighted sum = (0 x 4) + (4 x 2) + (-1) = 7.
Learning from labelled data is called
(d) supervised learning Supervised learning learns from examples with the correct labels.
The function that decides whether a neuron should 'fire' is the
(c) activation function The activation function decides the neuron's output.
A neuron has the inputs x1 = 5 with weight w1 = 2, x2 = 4 with weight w2 = 4 and bias 3. It gives the output 1 if the weighted sum is 30 or more, otherwise 0. What is the output?
(c) 0 Weighted sum = (5 x 2) + (4 x 4) + (3) = 29. It is less than the threshold 30.
A neuron has the inputs x1 = 2 with weight w1 = -2, x2 = 3 with weight w2 = 3 and bias -2. It gives the output 1 if the weighted sum is 5 or more, otherwise 0. What is the output?
(b) 0 Weighted sum = (2 x -2) + (3 x 3) + (-2) = 3. It is less than the threshold 5.
A neuron has the inputs x1 = 5 with weight w1 = 3, x2 = 4 with weight w2 = 3, x3 = 4 with weight w3 = 4. The bias is 3. What is the weighted sum (with bias)?
(a) 46 Weighted sum = (5 x 3) + (4 x 3) + (4 x 4) + (3) = 46.
A model in which the developer writes the rules is called
(b) rule-based In a rule-based model the rules are written by people.
The numbers that tell how important each input is to a neuron are called
(d) weights Weights show the importance of each input.
A neural network with many hidden layers is the basis of
(c) deep learning Deep learning uses neural networks with many hidden layers.
A supermarket studies bills to find items that are bought together. This is an example of
(c) Unsupervised learning Finding items bought together without labels is association (unsupervised learning).
A neuron has the inputs x1 = 1 with weight w1 = 3, x2 = 0 with weight w2 = -1, x3 = 0 with weight w3 = 1. The bias is 1. What is the weighted sum (with bias)?
(b) 4 Weighted sum = (1 x 3) + (0 x -1) + (0 x 1) + (1) = 4.
A neuron has the inputs x1 = 2 with weight w1 = 2, x2 = 2 with weight w2 = 3. The bias is 1. What is the weighted sum (with bias)?
(b) 11 Weighted sum = (2 x 2) + (2 x 3) + (1) = 11.
In an artificial neural network, the layer that receives the data is the
(c) input layer The input layer takes in the data.
In a data set of houses, 'area' and 'number of rooms' used to predict the price are
(d) features The inputs used to make a prediction are features.
Classification and regression are types of
(d) supervised learning Both learn from labelled data, so they are supervised learning.
Grouping customers with similar buying habits, without any labels, is
(d) clustering Making groups of similar items without labels is clustering.
A neuron has the inputs x1 = 3 with weight w1 = -2, x2 = 0 with weight w2 = -1, x3 = 5 with weight w3 = -2 and bias 1. It gives the output 1 if the weighted sum is -16 or more, otherwise 0. What is the output?
(c) 1 Weighted sum = (3 x -2) + (0 x -1) + (5 x -2) + (1) = -15. It is not less than the threshold -16.
A neuron has the inputs x1 = 1 with weight w1 = 4, x2 = 3 with weight w2 = 1, x3 = 1 with weight w3 = 4. The bias is -2. What is the weighted sum (with bias)?
(c) 9 Weighted sum = (1 x 4) + (3 x 1) + (1 x 4) + (-2) = 9.
"People who buy bread also buy butter." Finding such rules is called
(d) association Finding 'if this, then that' buying rules is association.
An app learns to recognise handwritten digits from 60,000 images, each tagged with the correct digit. This is an example of
(c) Supervised learning Each image has the correct label, so it is supervised learning.
Predicting whether an e-mail is 'spam' or 'not spam' is a problem of
(a) classification Predicting one of a fixed set of classes is classification.
A neuron has the inputs x1 = 0 with weight w1 = 3, x2 = 1 with weight w2 = 1. The bias is 1. What is the weighted sum (with bias)?
(a) 2 Weighted sum = (0 x 3) + (1 x 1) + (1) = 2.
Finding the one unusual transaction among thousands of normal ones is
(a) anomaly detection Spotting rare, unusual data is anomaly detection.
Which type of neural network is used mostly for images?
(b) Convolutional Neural Network (CNN) CNNs (Convolutional Neural Networks) are built for images.
The data used to teach a model is called
(a) training data Training data is used to teach the model.
A weather model is trained on past years' data with the recorded rainfall to predict rainfall. This is an example of
(d) Supervised learning The past data has the correct answers (recorded rainfall), so it is supervised learning.
A robot vacuum gets +1 for every clean tile and -1 for every bump, and improves its path. This is an example of
(a) Reinforcement learning It learns from rewards and penalties: reinforcement learning.
A neuron has the inputs x1 = 0 with weight w1 = -2, x2 = 3 with weight w2 = 4. The bias is -3. What is the weighted sum (with bias)?
(b) 9 Weighted sum = (0 x -2) + (3 x 4) + (-3) = 9.
A neuron has the inputs x1 = 3 with weight w1 = -1, x2 = 0 with weight w2 = 1. The bias is -1. What is the weighted sum (with bias)?
(d) -4 Weighted sum = (3 x -1) + (0 x 1) + (-1) = -4.
Which statement about AI, ML and DL is correct?
(d) DL is a part of ML, and ML is a part of AI DL ⊂ ML ⊂ AI: deep learning is a part of machine learning, which is a part of AI.
A dog is trained by giving a treat for every correct action. In AI this idea is used in
(b) reinforcement learning Rewarding correct actions is the idea behind reinforcement learning.
A program gives a discount if the bill is above 5,000 rupees, as written by the programmer. This is an example of
(b) Rule-based approach The rule is written by the programmer, so it is rule-based.
The extra value added to the weighted sum of a neuron is called the
(c) bias Bias is added to the weighted sum to shift the output.
A chatbot that answers only from a fixed list of 'if this, then that' replies follows the
(c) rule-based approach Fixed if-then replies are a rule-based approach.
In a table used for training, the column that the model has to predict is the
(b) label The target column is the label.
A model that finds the patterns from data on its own is called
(d) learning-based A learning-based model learns the patterns from data.
A neuron has the inputs x1 = 2 with weight w1 = -1, x2 = 3 with weight w2 = 1, x3 = 2 with weight w3 = 1. The bias is -1. What is the weighted sum (with bias)?
(b) 2 Weighted sum = (2 x -1) + (3 x 1) + (2 x 1) + (-1) = 2.
What does a high F1 score tell us about a model?
(c) Both its precision and its recall are good F1 is high only when both precision and recall are high.
A flood-warning AI gave an alarm, but no flood came. This outcome is a
(a) False Positive The prediction was positive (flood) and it was wrong.
Which is the correct meaning of precision?
(b) Out of all the positive predictions, the share that is actually positive Precision = TP / (TP + FP): of all 'yes' predictions, how many were right.
Which statement about the train-test split is INCORRECT?
(c) The model should be tested on the same data on which it was trained Testing on training data hides overfitting; the model must be tested on unseen data.
A model gave these results on the test data: TP = 24, FP = 24, FN = 8 and TN = 44. What is its precision?
(b) 0.50 Precision = TP / (TP + FP) = 24 / 48.
Checking how well a model works, using metrics, is called
(a) model evaluation Model evaluation uses metrics such as accuracy, precision, recall and F1.
A model gave these results on the test data: TP = 24, FP = 26, FN = 8 and TN = 42. What is its recall?
(d) 0.75 Recall = TP / (TP + FN) = 24 / 32.
Which is an ethical concern in model evaluation?
(a) The model may be biased against a group A model that works poorly for one group is biased and unfair.
A spam filter marked a genuine mail as spam. This outcome is a
(a) False Positive Predicted spam (positive) but really genuine: False Positive.
A model gave these results on the test data: TP = 26, FP = 26, FN = 14 and TN = 34. What is its accuracy?
(c) 60% Accuracy = (TP + TN) / Total = 60 / 100.
A model made 320 correct predictions out of 400 test cases. What is its accuracy?
(d) 80% Accuracy = correct predictions / total predictions = 320 / 400.
A rain-prediction app said 'rain', and it rained. This outcome is a
(a) True Positive Predicted yes and it was yes: True Positive.
A model gave these results on the test data: TP = 21, FP = 14, FN = 7 and TN = 8. What is its recall?
(c) 0.75 Recall = TP / (TP + FN) = 21 / 28.
A model predicted the weight of a parcel (in grams) as 360. The actual value was 400. What is the error rate?
(c) 10% Error = 40. Error rate = error / actual value = 40 / 400 = 10%.
A data set of 400 records is split 70:30 into training and testing data. How many records are used for testing?
(d) 120 Testing part = 30% of 400.
A test said 'not infected', but the person was infected. This outcome is a
(d) False Negative Predicted no but really yes: False Negative.
A model gave these results on the test data: TP = 54, FP = 6, FN = 18 and TN = 22. What is its recall?
(c) 0.75 Recall = TP / (TP + FN) = 54 / 72.
A spam filter must not send genuine mail to the spam folder. Which metric is most important?
(b) Precision A genuine mail marked as spam is a False Positive; precision measures it.
Out of 100 mails, 95 are genuine. A model marks every mail as genuine. Its accuracy is 95%, but it is useless because
(c) it never catches any spam (recall for spam is 0) Accuracy is 95/100, but it finds 0 of the 5 spam mails, so recall for spam is 0.
A model made 320 correct predictions out of 400 test cases. What is its accuracy?
(b) 80% Accuracy = correct predictions / total predictions = 320 / 400.
A model gave these results on the test data: TP = 44, FP = 44, FN = 36 and TN = 76. What is its precision?
(a) 0.50 Precision = TP / (TP + FP) = 44 / 88.
A model gave these results on the test data: TP = 60, FP = 40, FN = 20 and TN = 80. What is its accuracy?
(b) 70% Accuracy = (TP + TN) / Total = 140 / 200.
A spam filter let a genuine mail go to the inbox (positive class = spam). This outcome is a
(c) True Negative Predicted not spam and it was not spam: True Negative.
A model gave these results on the test data: TP = 21, FP = 14, FN = 4 and TN = 11. What is its precision?
(c) 0.60 Precision = TP / (TP + FP) = 21 / 35.
A model predicted the number of visitors to a fair as 360. The actual value was 400. What is the accuracy of this prediction?
(c) 90% Error = 40. Error rate = 40 / 400 = 10%. Accuracy = 100% - 10% = 90%.
Which formula gives recall?
(b) TP / (TP + FN) Recall = TP / (TP + FN): of all real positives, how many were found.
A model predicted the runs scored by a team as 736. The actual value was 800. What is the error rate?
(c) 8% Error = 64. Error rate = error / actual value = 64 / 800 = 8%.
A model gave these results on the test data: TP = 15, FP = 5, FN = 10 and TN = 20. What is its accuracy?
(c) 70% Accuracy = (TP + TN) / Total = 35 / 50.
A model gave these results on the test data: TP = 28, FP = 22, FN = 28 and TN = 22. What is its accuracy?
(d) 50% Accuracy = (TP + TN) / Total = 50 / 100.
The difference between the actual value and the predicted value is called
(c) error Error = actual value − predicted value.
A model gave these results on the test data: TP = 16, FP = 16, FN = 24 and TN = 44. What is its accuracy?
(d) 60% Accuracy = (TP + TN) / Total = 60 / 100.
A model gave these results on the test data: TP = 26, FP = 14, FN = 26 and TN = 34. What is its precision?
(d) 0.65 Precision = TP / (TP + FP) = 26 / 40.
A model made 800 correct predictions out of 1000 test cases. What is its accuracy?
(a) 80% Accuracy = correct predictions / total predictions = 800 / 1000.
A test for a serious disease must not miss any sick person. Which metric is most important?
(d) Recall Missing a sick person is a False Negative; recall measures it.
A model predicted the number of visitors to a fair as 540. The actual value was 500. What is the accuracy of this prediction?
(a) 92% Error = 40. Error rate = 40 / 500 = 8%. Accuracy = 100% - 8% = 92%.
Which formula gives accuracy?
(a) (TP + TN) / (TP + TN + FP + FN) Accuracy = correct predictions / all predictions = (TP + TN) / (TP + TN + FP + FN).
A model gives very good results on training data but poor results on new data. This is called
(b) overfitting It has memorised the training data instead of learning general patterns: overfitting.
A model gave these results on the test data: TP = 52, FP = 52, FN = 48 and TN = 48. What is its precision?
(b) 0.50 Precision = TP / (TP + FP) = 52 / 104.
A model predicted the weight of a parcel (in grams) as 960. The actual value was 800. What is the error rate?
(c) 20% Error = 160. Error rate = error / actual value = 160 / 800 = 20%.
A data set of 200 records is split 70:30 into training and testing data. How many records are used for testing?
(d) 60 Testing part = 30% of 200.
A face-unlock phone opened for a stranger (positive class = owner). This outcome is a
(b) False Positive Predicted owner (positive) but really a stranger: False Positive.
A table that compares the predictions of a model with the reality is a
(b) confusion matrix A confusion matrix shows TP, TN, FP and FN.
A smoke alarm did not ring, but there was a fire. This outcome is a
(a) False Negative Predicted no fire but there was a fire: False Negative.
A fraud detector flagged a genuine payment as fraud. This outcome is a
(a) False Positive Predicted fraud (positive) but really genuine: False Positive.
A model predicted the runs scored by a team as 525. The actual value was 500. What is the error rate?
(d) 5% Error = 25. Error rate = error / actual value = 25 / 500 = 5%.
An 80:20 train-test split of 500 records gives how many records for testing?
(d) 100 20% of 500 = 100 records for testing (400 for training).
A model gave these results on the test data: TP = 12, FP = 4, FN = 3 and TN = 31. What is its recall?
(d) 0.80 Recall = TP / (TP + FN) = 12 / 15.
A smoke alarm rang, and there was really a fire. This outcome is a
(c) True Positive Predicted fire and there was a fire: True Positive.
Which formula gives the F1 score?
(b) 2 x Precision x Recall / (Precision + Recall) F1 = 2 × Precision × Recall / (Precision + Recall), the harmonic mean.
The model predicted 'No' and the reality is 'No'. This is a
(a) True Negative Correct 'no': True Negative.
A plagiarism checker passed an essay that was really copied (positive class = copied). This outcome is a
(b) False Negative Predicted not copied but it was copied: False Negative.
The highest possible value of the F1 score is
(d) 1 F1 lies between 0 and 1; 1 means perfect precision and recall.
A model predicted the price of a bicycle (in rupees) as 224. The actual value was 200. What is the accuracy of this prediction?
(c) 88% Error = 24. Error rate = 24 / 200 = 12%. Accuracy = 100% - 12% = 88%.
A model gave these results on the test data: TP = 84, FP = 16, FN = 28 and TN = 72. What is its accuracy?
(a) 78% Accuracy = (TP + TN) / Total = 156 / 200.
A model gave these results on the test data: TP = 18, FP = 6, FN = 22 and TN = 54. What is its precision?
(b) 0.75 Precision = TP / (TP + FP) = 18 / 24.
A data set of 500 records is split 70:30 into training and testing data. How many records are used for testing?
(b) 150 Testing part = 30% of 500.
When a model's working and results can be explained and checked by people, the model is
(c) transparent Such a model is transparent (explainable).
A model gave these results on the test data: TP = 24, FP = 26, FN = 8 and TN = 42. What is its recall?
(b) 0.75 Recall = TP / (TP + FN) = 24 / 32.
A model gave these results on the test data: TP = 36, FP = 4, FN = 44 and TN = 116. What is its accuracy?
(a) 76% Accuracy = (TP + TN) / Total = 152 / 200.
If the accuracy of a model is 92%, its error rate is
(b) 8% Error rate = 100% − accuracy = 8%.
A model gave these results on the test data: TP = 9, FP = 6, FN = 11 and TN = 24. What is its recall?
(c) 0.45 Recall = TP / (TP + FN) = 9 / 20.
A metal detector stayed silent, and the bag had no metal. This outcome is a
(a) True Negative Predicted no and it was no: True Negative.
A model gave these results on the test data: TP = 112, FP = 28, FN = 48 and TN = 12. What is its recall?
(c) 0.70 Recall = TP / (TP + FN) = 112 / 160.
A model gave these results on the test data: TP = 48, FP = 16, FN = 12 and TN = 124. What is its precision?
(d) 0.75 Precision = TP / (TP + FP) = 48 / 64.
A model gave these results on the test data: TP = 18, FP = 18, FN = 12 and TN = 52. What is its accuracy?
(b) 70% Accuracy = (TP + TN) / Total = 70 / 100.
The model predicted 'Yes' and the reality is 'Yes'. This is a
(a) True Positive Correct 'yes': True Positive.
A model gave these results on the test data: TP = 26, FP = 14, FN = 24 and TN = 36. What is its precision?
(c) 0.65 Precision = TP / (TP + FP) = 26 / 40.
A model gave these results on the test data: TP = 9, FP = 3, FN = 9 and TN = 29. What is its recall?
(c) 0.50 Recall = TP / (TP + FN) = 9 / 18.
A model predicted the price of a bicycle (in rupees) as 550. The actual value was 500. What is the accuracy of this prediction?
(a) 90% Error = 50. Error rate = 50 / 500 = 10%. Accuracy = 100% - 10% = 90%.
In a confusion matrix, the correct predictions are
(d) TP and TN TP and TN are correct; FP and FN are errors.
The model predicted 'No' but the reality is 'Yes'. This is a
(d) False Negative Missed a real 'yes': False Negative.
A model that gives 100% accuracy on the data it was trained on is
(b) not proved good, because it has only memorised that data It may only have memorised that data; it must be tested on new data.
A model gave these results on the test data: TP = 18, FP = 7, FN = 12 and TN = 13. What is its recall?
(a) 0.60 Recall = TP / (TP + FN) = 18 / 30.
A model gave these results on the test data: TP = 36, FP = 24, FN = 12 and TN = 28. What is its accuracy?
(b) 64% Accuracy = (TP + TN) / Total = 64 / 100.
A model gave these results on the test data: TP = 18, FP = 2, FN = 18 and TN = 62. What is its precision?
(d) 0.90 Precision = TP / (TP + FP) = 18 / 20.
Finding the objects in an image or video and marking where each one is, is called
(b) object detection Object detection finds many objects and draws a box around each.
Which is a use of Computer Vision in online shopping?
(c) Searching for a product by uploading its photo Visual search finds products from a photo.
Marking the exact outline (pixels) of every separate object in an image is
(a) instance segmentation Instance segmentation marks the exact pixels of each separate object.
Google Lens identifying a plant from its photo is an application of
(c) Computer Vision It understands an image, so it is Computer Vision.
Classifying the single object in an image and also marking its place with a box is
(c) classification + localisation Saying what the object is (classification) and where it is (localisation).
The process of sliding a small kernel over an image to bring out features such as edges is
(c) convolution Convolution multiplies the kernel with each patch of pixels and adds them up.
Which is NOT an application of Computer Vision?
(a) Language translation Language translation works with text, so it is NLP, not computer vision.
The AI domain that helps machines to see and understand images and videos is
(b) Computer Vision Computer Vision lets machines understand images and videos.
The kernel used in convolution is a
(d) small matrix of numbers A kernel (filter) is a small matrix of numbers, such as 3 × 3.
In a grayscale image, a pixel value of 0 means
(b) black 0 is the darkest value: black (255 is white).
Which has the highest resolution?
(b) 1920 x 1080 1920 × 1080 has the most pixels (about 2 million).
In a colour image, each pixel stores
(c) three values, one each for red, green and blue Each pixel stores red, green and blue values.
An image is stored in the computer as
(c) numbers Every pixel is stored as numbers.
Which features of an image are most useful for a machine to recognise objects?
(c) Corners and edges Corners and edges are distinctive, so they help to recognise objects.
In an RGB image, the pixel (255, 255, 255) is
(b) white All three colours at full brightness give white.
A self-driving car finding pedestrians and traffic lights on the road uses
(b) Computer Vision Finding objects in camera images is Computer Vision.
Which tool can be used to train an image model without writing code?
(d) Teachable Machine Teachable Machine (by Google) trains models in the browser without code.
The number of pixels in an image (width x height) is called its
(b) resolution Resolution is width × height in pixels.
An RGB image has the resolution 4 x 3. How many values (numbers) are stored for the whole image?
(c) 36 Each of the 12 pixels stores 3 values: red, green and blue.
An RGB image has the resolution 640 x 480. How many values (numbers) are stored for the whole image?
(c) 921,600 Each of the 307,200 pixels stores 3 values: red, green and blue.
How many shades can one pixel of an 8-bit grayscale image show?
(a) 256 8 bits give 2⁸ = 256 shades (0 to 255).
How many channels does an RGB image have?
(c) 3 Three channels: Red, Green and Blue.
How many pixels are there in an image of resolution 28 x 28?
(b) 784 Pixels = width x height.
The smallest unit of a digital image is a
(c) pixel A pixel (picture element) is the smallest unit.
How many pixels are there in an image of resolution 8 x 8?
(b) 64 Pixels = width x height.
Which task works on an image that has many objects?
(a) Object detection Object detection finds many objects in one image.
A camera that reads number plates at a toll plaza uses
(c) Computer Vision Reading plates from images is Computer Vision.
The range of pixel values in an 8-bit image is
(a) 0 to 255 8 bits store 0 to 255.
An 8-bit grayscale image has the resolution 64 x 64. How many bytes are needed to store its pixel values?
(a) 4,096 One byte for each pixel: 64 x 64.
How many pixels are there in an image of resolution 10 x 5?
(d) 50 Pixels = width x height.
In an RGB image, the pixel (255, 0, 0) is
(b) red Full red, no green, no blue: red.
A megapixel is equal to about
(a) one million pixels Mega = million, so 1 megapixel ≈ one million pixels.
Telling only whether a photo shows a 'cat' or a 'dog' is
(d) image classification Giving one label to the whole image is image classification.
An 8-bit grayscale image has the resolution 28 x 28. How many bytes are needed to store its pixel values?
(c) 784 One byte for each pixel: 28 x 28.
How many pixels are there in an image of resolution 10 x 5?
(c) 50 Pixels = width x height.
In a CNN, which layer makes the feature map smaller while keeping the important features?
(b) Pooling layer Pooling (e.g. max pooling) shrinks the feature map and keeps the strongest values.
The last layer of a CNN that gives the final class is the
(a) fully connected layer The fully connected layer combines the features and gives the class.
The word 'pixel' comes from
(b) picture element pixel = picture element.
An 8-bit grayscale image has the resolution 4 x 3. How many bytes are needed to store its pixel values?
(d) 12 One byte for each pixel: 4 x 3.
In a grayscale image, a pixel value of 255 means
(b) white 255 is the brightest value: white.
RGB stands for
(a) Red, Green, Blue R = Red, G = Green, B = Blue.
An RGB image has the resolution 28 x 28. How many values (numbers) are stored for the whole image?
(b) 2,352 Each of the 784 pixels stores 3 values: red, green and blue.
The base form (lemma) of the word 'better' is
(b) good Lemmatisation gives a meaningful dictionary word.
What is the main goal of keyword extraction?
(c) To find the most important words or phrases of a text Keyword extraction picks out the most important words or phrases.
An NLP application that sorts news articles into sports, politics and business is
(a) text classification Putting texts into categories is text classification.
A corpus has three documents. D1: Trees give us oxygen. / D2: We must plant trees. / D3: Plant trees and save the earth. The word 'trees' occurs once in Document 1. What is its TF-IDF value in Document 1? (log 3 = 0.477, log 1.5 = 0.176, log 1 = 0)
(b) 0 TF = 1, IDF = 3/3, so TF-IDF = 1 x log(3/3) = 0.
Cutting the affixes of a word to get its root, even if the root is not a meaningful word, is
(b) stemming Stemming just cuts endings, e.g. 'studies' → 'studi'.
The Bag of Words model gives us
(a) the vocabulary of the corpus and the frequency of each word BoW gives the vocabulary and how many times each word occurs.
The stop words a, an, the, is, are, to, and, of, in, on, at, it, my, we, he, she, with are removed from the sentence "Delhi is the capital of India." (punctuation is also removed). How many words are left?
(c) 3 Words left: delhi, capital, india.
In TF-IDF, IDF stands for
(a) Inverse Document Frequency IDF = Inverse Document Frequency.
A corpus has three documents. D1: Yoga keeps the body fit. / D2: A fit body has a fit mind. / D3: Yoga keeps the mind calm. What is the document frequency of the word 'calm'?
(b) 1 Document frequency = number of documents that contain the word.
A chatbot that learns from data and improves with use is a
(c) smart bot A smart bot uses AI to learn and improve.
A corpus has three documents. D1: The bus was late today. / D2: I missed the bus. / D3: I was late for school today. The word 'the' occurs once in Document 1. What is its TF-IDF value in Document 1? (log 3 = 0.477, log 1.5 = 0.176, log 1 = 0)
(c) 0.176 TF = 1, IDF = 3/2, so TF-IDF = 1 x log(3/2) = 0.176.
Converting all the text to lower case in normalisation is done so that
(d) 'Hello' and 'hello' are treated as the same word Lower case makes 'Hello' and 'hello' the same token.
Dividing a sentence into words, numbers and special characters is called
(b) tokenisation Tokenisation splits a sentence into tokens.
How many tokens are there in the sentence: "Wow! We won the match."
(c) 7 Every word, number and punctuation mark is one token: Wow | ! | We | won | the | match | ..
What is the lemma of the word 'feet'?
(c) foot Lemmatisation gives the meaningful base word.
Which of these is a smart bot?
(b) A voice assistant such as Alexa or Google Assistant Alexa and Google Assistant understand varied questions and learn: smart bots.
The number of documents in which a word occurs is its
(d) document frequency Document frequency counts the documents that contain the word.
Which is NOT an application of NLP?
(d) Face recognition Face recognition works on images, so it is Computer Vision.
The stop words a, an, the, is, are, to, and, of, in, on, at, it, my, we, he, she, with are removed from the sentence "Practice makes a man perfect!" (punctuation is also removed). How many words are left?
(d) 4 Words left: practice, makes, man, perfect.
Finding the meaning of words and sentences is
(a) semantic analysis Semantic analysis is about meaning.
The stop words a, an, the, is, are, to, and, of, in, on, at, it, my, we, he, she, with are removed from the sentence "I love my school!" (punctuation is also removed). How many words are left?
(a) 2 Words left: love, school.
A corpus has three documents. D1: We love our school. / D2: Our school has a big library. / D3: We read books in the library. After changing to lower case and removing punctuation (no word is dropped), how many words are there in the vocabulary?
(a) 12 Each different word is counted once: we, love, our, school, has, a, big, library, read, books, in, the.
The stop words a, an, the, is, are, to, and, of, in, on, at, it, my, we, he, she, with are removed from the sentence "My name is Asha." (punctuation is also removed). How many words are left?
(c) 2 Words left: name, asha.
How many tokens are there in the sentence: "The cat sat on the mat."
(b) 7 Every word, number and punctuation mark is one token: The | cat | sat | on | the | mat | ..
Finding whether a product review is positive, negative or neutral is
(b) sentiment analysis Sentiment analysis finds the feeling (opinion) in a text.
A corpus has three documents. D1: Divya went to the market. / D2: The market was closed. / D3: Divya came back home. After changing to lower case and removing punctuation (no word is dropped), how many words are there in the vocabulary?
(d) 10 Each different word is counted once: divya, went, to, the, market, was, closed, came, back, home.
What is the lemma of the word 'caring'?
(c) care Lemmatisation gives the meaningful base word.
The stop words a, an, the, is, are, to, and, of, in, on, at, it, my, we, he, she, with are removed from the sentence "The sun rises in the east." (punctuation is also removed). How many words are left?
(a) 3 Words left: sun, rises, east.
How many tokens are there in the sentence: "Is it raining today?"
(c) 5 Every word, number and punctuation mark is one token: Is | it | raining | today | ?.
Words such as 'a', 'an', 'the', 'is', 'to', which occur very often but add little meaning, are
(a) stop words Stop words are very common words that carry little meaning.
The stop words a, an, the, is, are, to, and, of, in, on, at, it, my, we, he, she, with are removed from the sentence "Hello, how are you?" (punctuation is also removed). How many words are left?
(a) 3 Words left: hello, how, you.
"Chickens feed extravagantly while the moon drinks tea." This sentence has
(c) correct syntax but no sensible meaning The grammar is correct but the meaning makes no sense: correct syntax, wrong semantics.
The stop words a, an, the, is, are, to, and, of, in, on, at, it, my, we, he, she, with are removed from the sentence "She bought 3 pens and 2 books." (punctuation is also removed). How many words are left?
(a) 5 Words left: bought, 3, pens, 2, books.
What is the lemma of the word 'children'?
(a) child Lemmatisation gives the meaningful base word.
The stop words a, an, the, is, are, to, and, of, in, on, at, it, my, we, he, she, with are removed from the sentence "Save water, save life." (punctuation is also removed). How many words are left?
(a) 4 Words left: save, water, save, life.
How many tokens are there in the sentence: "We are learning Python at school."
(b) 7 Every word, number and punctuation mark is one token: We | are | learning | Python | at | school | ..
The AI domain that helps machines to understand human language is
(c) Natural Language Processing NLP (Natural Language Processing) deals with human language.
A corpus has three documents. D1: Robots can help doctors. / D2: Doctors help sick people. / D3: Robots can not feel. What is the document frequency of the word 'feel'?
(b) 1 Document frequency = number of documents that contain the word.
How many tokens are there in the sentence: "AI is changing the world, right?"
(d) 8 Every word, number and punctuation mark is one token: AI | is | changing | the | world | , | right | ?.
A corpus has three documents. D1: The sun is hot. / D2: The moon is cool. / D3: The sun and the moon are in the sky. What is the document frequency of the word 'moon'?
(a) 2 Document frequency = number of documents that contain the word.
The stem of 'studies' is 'studi' and the lemma is 'study'. This shows that
(a) lemmatisation gives a meaningful word but stemming may not Lemmatisation gives a real word (study); stemming may not (studi).
What is the lemma of the word 'went'?
(b) go Lemmatisation gives the meaningful base word.
How many tokens are there in the sentence: "Reading makes a full man."
(b) 6 Every word, number and punctuation mark is one token: Reading | makes | a | full | man | ..
A corpus has three documents. D1: Data is the new oil. / D2: AI needs data. / D3: AI is the future. The word 'data' occurs once in Document 1. What is its TF-IDF value in Document 1? (log 3 = 0.477, log 1.5 = 0.176, log 1 = 0)
(b) 0.176 TF = 1, IDF = 3/2, so TF-IDF = 1 x log(3/2) = 0.176.
A corpus has three documents. D1: I like mango juice. / D2: Mango is the king of fruits. / D3: I like fruits. What is the document frequency of the word 'of'?
(d) 1 Document frequency = number of documents that contain the word.
In TF-IDF, TF stands for
(b) Term Frequency TF = Term Frequency, how often a word occurs in a document.
Dividing the corpus into sentences is called
(b) sentence segmentation Sentence segmentation splits text into sentences.
The whole collection of text documents used in an NLP project is called the
(b) corpus The corpus is the whole collection of text.
What is the lemma of the word 'studies'?
(a) study Lemmatisation gives the meaningful base word.
A corpus has three documents. D1: Solar panels save power. / D2: We save power at home. / D3: Solar power is clean. The word 'power' occurs once in Document 1. What is its TF-IDF value in Document 1? (log 3 = 0.477, log 1.5 = 0.176, log 1 = 0)
(a) 0 TF = 1, IDF = 3/3, so TF-IDF = 1 x log(3/3) = 0.
Checking the grammar and arrangement of words in a sentence is
(d) syntactic analysis (syntax) Syntactic analysis checks grammar and word order.
Which step comes first in text normalisation?
(c) Sentence segmentation Sentence segmentation, then tokenisation, stop-word removal, lower case, stemming/lemmatisation.
How many tokens are there in the sentence: "Raj and Vicky are best friends."
(c) 7 Every word, number and punctuation mark is one token: Raj | and | Vicky | are | best | friends | ..
A word that occurs in every document of the corpus has a TF-IDF value that is
(b) zero or very low log(N/N) = log 1 = 0.
A corpus has three documents. D1: India won the final match. / D2: The final match was in Mumbai. / D3: India plays well. After changing to lower case and removing punctuation (no word is dropped), how many words are there in the vocabulary?
(c) 10 Each different word is counted once: india, won, the, final, match, was, in, mumbai, plays, well.
How many tokens are there in the sentence: "He is playing in the park with a ball."
(b) 10 Every word, number and punctuation mark is one token: He | is | playing | in | the | park | with | a | ball | ..
A chatbot that works on a fixed script and cannot handle unexpected questions is a
(c) script bot A script bot follows fixed rules only.
In which word does the meaning change with the sentence: 'bank of the river' and 'money in the bank'?
(b) bank The same word can have different meanings; this makes human language hard for machines.
What is the lemma of the word 'running'?
(d) run Lemmatisation gives the meaningful base word.
A corpus has three documents. D1: Ravi plays football daily. / D2: Football is a team game. / D3: Ravi loves his team. After changing to lower case and removing punctuation (no word is dropped), how many words are there in the vocabulary?
(c) 10 Each different word is counted once: ravi, plays, football, daily, is, a, team, game, loves, his.
Statement 1: Sector-based frameworks are made for one field, such as health care.
Statement 2: Value-based frameworks look at the values behind a decision, such as rights, utility and virtue.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: Bioethics is a value-based framework.
Statement 2: Rights-based, utility-based and virtue-based are value-based frameworks.
(d) Statement 1 is false and Statement 2 is true. Statement 1 is false; Statement 2 is true.
Statement 1: Problem Scoping is the last stage of the AI Project Cycle.
Statement 2: Deployment comes before Data Acquisition.
(b) Both the statements are false. Statement 1 is false; Statement 2 is false.
Statement 1: The 4Ws canvas is used in Problem Scoping.
Statement 2: The 4Ws are Who, What, Where and Why.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: Data Exploration helps us to see trends and patterns in the data.
Statement 2: Graphs and charts are not used in Data Exploration.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: Computer Vision works with images and videos.
Statement 2: Natural Language Processing works only with numbers in tables.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: 'Do not harm' is a principle of bioethics.
Statement 2: 'Give justice' is a principle of bioethics.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: An ethical framework helps to avoid unintended harm from an AI system.
Statement 2: Ethics is needed only after the AI system has caused harm.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: A utility-based framework looks for the greatest good for the greatest number.
Statement 2: A rights-based framework ignores human dignity.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: Evaluation is done before Modelling.
Statement 2: Data Acquisition is done after Problem Scoping.
(d) Statement 1 is false and Statement 2 is true. Statement 1 is false; Statement 2 is true.
Statement 1: Machine Learning is a subset of Artificial Intelligence.
Statement 2: Deep Learning is a subset of Machine Learning.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: Supervised learning uses labelled data.
Statement 2: Clustering is a type of supervised learning.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: Regression predicts a continuous number.
Statement 2: Classification predicts a category.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: In a rule-based model, the machine finds the rules itself.
Statement 2: In a learning-based model, the developer writes all the rules.
(b) Both the statements are false. Statement 1 is false; Statement 2 is false.
Statement 1: The hidden layers of a neural network do most of the processing.
Statement 2: The output layer receives the raw input data.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: Reinforcement learning uses rewards and penalties.
Statement 2: Association finds items that often occur together.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: Precision = TP / (TP + FP).
Statement 2: Recall = TP / (TP + FP).
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: A model should be tested on data it has not seen in training.
Statement 2: The testing data is usually larger than the training data.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: The F1 score is high only when both precision and recall are high.
Statement 2: The F1 score can be more than 1.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: A False Positive means the model predicted 'Yes' but the reality is 'No'.
Statement 2: A False Negative means the model predicted 'No' but the reality is 'Yes'.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: Accuracy is always the best metric, even when one class is very rare.
Statement 2: TP and TN are the wrong predictions of a model.
(b) Both the statements are false. Statement 1 is false; Statement 2 is false.
Statement 1: When False Negatives are costly, recall is the important metric.
Statement 2: When False Positives are costly, precision is the important metric.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: A grayscale pixel has a value from 0 to 255.
Statement 2: An RGB image has three channels.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: Image classification tells where each object is with a box.
Statement 2: Object detection can find many objects in one image.
(d) Statement 1 is false and Statement 2 is true. Statement 1 is false; Statement 2 is true.
Statement 1: Resolution is the number of pixels in an image.
Statement 2: In a grayscale image, 0 stands for white.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: Instance segmentation marks the pixels of each separate object.
Statement 2: Computer Vision is used in face recognition.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: A pixel is the largest unit of a digital image.
Statement 2: A 1 megapixel image has about one thousand pixels.
(b) Both the statements are false. Statement 1 is false; Statement 2 is false.
Statement 1: Stemming always gives a meaningful word.
Statement 2: Lemmatisation gives a meaningful word.
(d) Statement 1 is false and Statement 2 is true. Statement 1 is false; Statement 2 is true.
Statement 1: Tokenisation divides a sentence into tokens.
Statement 2: A punctuation mark is also counted as a token.
(a) Both the statements are true. Statement 1 is true; Statement 2 is true.
Statement 1: Stop words such as 'the' and 'is' occur very often in a corpus.
Statement 2: Stop words have the highest TF-IDF values.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: A script bot learns from data on its own.
Statement 2: A smart bot works only on a fixed script.
(b) Both the statements are false. Statement 1 is false; Statement 2 is false.
Statement 1: Bag of Words gives the frequency of each word of the vocabulary.
Statement 2: Bag of Words keeps the order of the words of a sentence.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Statement 1: Sentiment analysis is an application of NLP.
Statement 2: Language translation is an application of Computer Vision.
(c) Statement 1 is true and Statement 2 is false. Statement 1 is true; Statement 2 is false.
Assertion (A): The 4Ws problem canvas is filled in the Problem Scoping stage.
Reason (R): The 4Ws help us to understand who has the problem, what it is, where it occurs and why it should be solved.
(a) Both A and R are true and R is the correct explanation of A. The 4Ws help to understand the problem (R), which is the job of Problem Scoping (A).
Assertion (A): Bioethics is a sector-based ethical framework.
Reason (R): Bioethics deals with ethical questions of health care and life sciences.
(a) Both A and R are true and R is the correct explanation of A. Bioethics deals with one sector, health care and life sciences (R), so it is sector-based (A).
Assertion (A): Data Exploration is done before Modelling.
Reason (R): Understanding patterns in the data helps in choosing a suitable model.
(a) Both A and R are true and R is the correct explanation of A. Patterns found in exploration help to choose the model (R), so it comes first (A).
Assertion (A): An AI system should be checked for bias before deployment.
Reason (R): A biased system can treat some groups of people unfairly.
(a) Both A and R are true and R is the correct explanation of A. A biased system treats some groups unfairly (R), so it must be checked first (A).
Assertion (A): Evaluation is the first stage of the AI Project Cycle.
Reason (R): A model has to be tested before it is deployed.
(d) A is false but R is true. Problem Scoping is the first stage.
Assertion (A): Deployment is done after Evaluation.
Reason (R): Data is collected in the Data Acquisition stage.
(b) Both A and R are true but R is not the correct explanation of A. Both are true, but R talks about data acquisition, which does not explain why deployment follows evaluation.
Assertion (A): A utility-based framework always protects the rights of every single person.
Reason (R): A utility-based framework looks for the greatest good of the greatest number.
(d) A is false but R is true. The good of the majority may still hurt a few people.
Assertion (A): Ethical frameworks are useful for AI developers.
Reason (R): They give a structured way to take morally right decisions.
(a) Both A and R are true and R is the correct explanation of A. They give a structured way to take right decisions (R), which makes them useful (A).
Assertion (A): Computer Vision is one of the domains of AI.
Reason (R): Computer Vision works on text documents only.
(c) A is true but R is false. A is true, but R is false: computer vision works on images and videos, not text.
Assertion (A): Data should be collected from reliable sources.
Reason (R): Wrong or biased data leads to a wrong or biased model.
(a) Both A and R are true and R is the correct explanation of A. Wrong or biased data gives a wrong or biased model (R), so sources must be reliable (A).
Assertion (A): Supervised learning needs labelled data.
Reason (R): Clustering is a type of supervised learning.
(c) A is true but R is false. Clustering is unsupervised learning.
Assertion (A): Predicting tomorrow's temperature is a regression problem.
Reason (R): Regression predicts a continuous numeric value.
(a) Both A and R are true and R is the correct explanation of A. Temperature is a continuous number and regression predicts numbers (R).
Assertion (A): Grouping news articles without any labels is clustering.
Reason (R): Unsupervised learning works on unlabelled data.
(a) Both A and R are true and R is the correct explanation of A. Unsupervised learning works without labels (R), and grouping is clustering (A).
Assertion (A): A neural network can learn from data.
Reason (R): During training, the weights and bias of its neurons are adjusted.
(a) Both A and R are true and R is the correct explanation of A. Training adjusts the weights and bias (R), which is how it learns (A).
Assertion (A): A rule-based model learns the rules from the data by itself.
Reason (R): In a learning-based model the machine finds the patterns from data.
(d) A is false but R is true. A is false: in a rule-based model people write the rules. R (about learning-based models) is true.
Assertion (A): Deep Learning is a subset of Machine Learning.
Reason (R): Deep Learning uses neural networks with many layers.
(b) Both A and R are true but R is not the correct explanation of A. Both are true, but R tells how deep learning works (many layers), not why it is a part of ML.
Assertion (A): A model must be tested on unseen data.
Reason (R): Testing on training data only shows what the model has memorised.
(a) Both A and R are true and R is the correct explanation of A. Testing on training data only shows memory (R), so unseen data is needed (A).
Assertion (A): Recall is the most important metric for a model that detects a serious disease.
Reason (R): A False Negative means a sick person is told that he or she is healthy.
(a) Both A and R are true and R is the correct explanation of A. A false negative sends a sick person home (R); recall counts how few such misses happen (A).
Assertion (A): Accuracy alone can be misleading.
Reason (R): When one class is very rare, a model can get high accuracy by always predicting the common class.
(a) Both A and R are true and R is the correct explanation of A. With a rare class, always predicting the common class gives high accuracy (R), so accuracy alone misleads (A).
Assertion (A): The F1 score of a model can be 1.5.
Reason (R): The F1 score combines precision and recall.
(d) A is false but R is true. F1 lies between 0 and 1.
Assertion (A): Precision is important for a spam filter.
Reason (R): Precision = TP / (TP + FN).
(c) A is true but R is false. Precision = TP / (TP + FP).
Assertion (A): A confusion matrix for two classes has four cells.
Reason (R): The four cells are TP, TN, FP and FN.
(a) Both A and R are true and R is the correct explanation of A. The four cells are TP, TN, FP and FN (R).
Assertion (A): Computer Vision systems can identify objects in images.
Reason (R): A machine sees an image as a grid of numbers called pixel values.
(b) Both A and R are true but R is not the correct explanation of A. Both are true, but R only says how images are stored; it does not explain how objects are identified.
Assertion (A): An RGB image needs more storage than a grayscale image of the same resolution.
Reason (R): Each pixel of an RGB image stores three values in place of one.
(a) Both A and R are true and R is the correct explanation of A. Each RGB pixel stores three values instead of one (R), so it needs about three times the space (A).
Assertion (A): Object detection is used in self-driving cars.
Reason (R): Object detection finds the objects in a scene and their locations.
(a) Both A and R are true and R is the correct explanation of A. It finds objects and their positions (R), which the car needs to drive safely (A).
Assertion (A): In a grayscale image, the pixel value 255 shows black.
Reason (R): Pixel values of an 8-bit image lie between 0 and 255.
(d) A is false but R is true. 255 is white and 0 is black.
Assertion (A): Stop words are usually removed during text normalisation.
Reason (R): Stop words occur very often but add little meaning to the corpus.
(a) Both A and R are true and R is the correct explanation of A. They occur often but add little meaning (R), so they are removed (A).
Assertion (A): Lemmatisation takes more time than stemming.
Reason (R): Lemmatisation makes sure that the root word is a meaningful word.
(a) Both A and R are true and R is the correct explanation of A. Lemmatisation must make sure the root is a real word (R), which needs more work (A).
Assertion (A): A word that occurs in every document has a high TF-IDF value.
Reason (R): IDF becomes smaller as the document frequency becomes larger.
(d) A is false but R is true. A is false: IDF = log(N/DF) is 0 when the word is in every document (R is true), so TF-IDF is low.
Assertion (A): A script bot cannot answer a question outside its script.
Reason (R): A script bot works on pre-written rules.
(a) Both A and R are true and R is the correct explanation of A. It works only on pre-written rules (R), so unexpected questions fail (A).
Assertion (A): Human language is difficult for a machine to understand.
Reason (R): The same word can have different meanings in different sentences.
(a) Both A and R are true and R is the correct explanation of A. One word can mean different things in different sentences (R), which makes language hard (A).
Assertion (A): Tokenisation is done before sentence segmentation.
Reason (R): A token can be a word, a number or a special character.
(d) A is false but R is true. Sentence segmentation comes first.
Assertion (A): Feedback completes the communication cycle.
Reason (R): Without feedback the sender cannot know whether the message was understood.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Specific feedback is more useful than non-specific feedback.
Reason (R): Specific feedback tells exactly what is good and what to improve.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Written communication gives immediate feedback.
Reason (R): Written communication provides a permanent record.
(d) A is false but R is true. Oral communication gives immediate feedback; written feedback is slow. R is true.
Assertion (A): Crossed arms and looking away show positive body language.
Reason (R): Positive body language includes smiling, nodding and eye contact.
(d) A is false but R is true. Crossed arms and looking away are negative body language, so A is false.
Assertion (A): "The letter was signed by the manager" is in the passive voice.
Reason (R): In the passive voice the receiver of the action comes first.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): A small amount of stress can improve performance.
Reason (R): Eustress motivates and improves focus.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): SMART goals have a deadline.
Reason (R): The T in SMART stands for Time-bound.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Procrastination helps in time management.
Reason (R): Procrastination means delaying work.
(d) A is false but R is true. Procrastination wastes time, so A is false; R is true.
Assertion (A): Talking to family and friends can reduce stress.
Reason (R): Sharing problems reduces worry and gives support.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Studying only to get a new phone is internal motivation.
Reason (R): Internal motivation comes from interest and enjoyment of the task.
(d) A is false but R is true. Studying for a phone is external motivation, so A is false; R is true.
Assertion (A): The operating system acts as an interface between the user and the hardware.
Reason (R): Application software such as a word processor runs on top of the operating system.
(b) Both A and R are true but R is not the correct explanation of A. Both are true, but R does not explain why the OS is an interface.
Assertion (A): A worm needs a user to run an infected file to spread.
Reason (R): A worm copies itself and spreads automatically through networks.
(d) A is false but R is true. A worm spreads by itself; A is false.
Assertion (A): A Trojan horse looks like a useful program.
Reason (R): A Trojan horse copies itself to other computers.
(c) A is true but R is false. A Trojan does not replicate itself, so R is false.
Assertion (A): Shift + Delete removes a file permanently.
Reason (R): Files deleted with Shift + Delete do not go to the Recycle Bin.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Regular backups protect data.
Reason (R): If the computer is damaged or infected, the backup copy can be restored.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Entrepreneurs take risks.
Reason (R): They invest money and effort without being sure of profit.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Innovation is a function of an entrepreneur.
Reason (R): Entrepreneurs create new products and better methods.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): A wage-employed person bears the risk of the business.
Reason (R): Wage employees get a fixed salary.
(d) A is false but R is true. The business risk is borne by the owner, so A is false; R is true.
Assertion (A): Failure means the end of an entrepreneur's career.
Reason (R): Many entrepreneurs learn from failure and succeed later.
(d) A is false but R is true. A is a myth; R is true.
Assertion (A): A business plan is useful for an entrepreneur.
Reason (R): It shows the product, customers, costs, price and expected profit.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): The SDGs were adopted by the United Nations in 2015.
Reason (R): They are to be achieved by 2030.
(b) Both A and R are true but R is not the correct explanation of A. Both are true, but R does not explain A.
Assertion (A): Rainwater harvesting supports sustainable development.
Reason (R): It conserves water for future use.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Organic farming harms soil health.
Reason (R): Organic farming avoids chemical fertilisers and pesticides.
(d) A is false but R is true. Organic farming improves soil health, so A is false.
Assertion (A): Sustainable development has three dimensions.
Reason (R): They are economic, social and environmental.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Assertion (A): Using public transport reduces air pollution.
Reason (R): Public transport carries more people with fewer vehicles.
(a) Both A and R are true and R is the correct explanation of A. Both statements are true, and R gives the reason why A is true.
Name the types of verbal communication, with one example of each.
(1) Oral (spoken) communication: face-to-face talk, phone call, speech. (2) Written communication: letter, e-mail, SMS, notice. Words are used in both.
What is the communication cycle? Name its elements.
It is the path a message takes from the sender to the receiver and back. Elements: sender, message, encoding, channel, receiver, decoding and feedback.
Write any four of the 7 Cs of effective communication.
Clear, Concise, Concrete, Correct, Coherent, Complete, Courteous. (Any four; one line on each, for example Concise: use only the words that are needed.)
What is non-verbal communication? Give two examples.
Sending a message without words, through expressions, gestures, posture, touch or eye contact. Examples: nodding the head for 'yes', a firm handshake, a smile.
What is feedback? Why is it important?
Feedback is the receiver's response to the sender's message. It shows whether the message was understood, and it helps the sender to improve.
Write two differences between verbal and non-verbal communication.
(1) Verbal communication uses words, spoken or written; non-verbal uses gestures, expressions and body language. (2) Verbal messages can be recorded exactly; non-verbal signals can be understood differently by different people.
Name any four barriers to effective communication.
Physical barriers (noise, distance), linguistic barriers (different language), interpersonal barriers (lack of trust or interest), organisational barriers (strict rules, hierarchy) and cultural barriers. (Any four.)
What is visual communication? Give two examples.
Sending a message through pictures, signs, symbols, charts and maps. Examples: traffic signs, the 'no smoking' symbol, a bar chart of results.
Write two ways to overcome barriers to communication.
(1) Use simple language and speak clearly; check that the listener has understood (take feedback). (2) Choose a quiet place and the right channel; respect the culture and feelings of the other person.
What are descriptive (specific) feedback and non-specific feedback? Give one example of each.
Specific feedback tells exactly what is good or what to improve: "Your introduction is clear; add one example." Non-specific feedback is general: "Good job."
Identify the subject, verb and object in the sentence: "The teacher praised the students." Which kind of sentence is it?
Subject: The teacher. Verb: praised. Object: the students. It is a declarative (statement) sentence.
What does SMART stand for in goal setting?
Specific, Measurable, Achievable, Realistic and Time bound. A SMART goal is clear, can be measured, is possible to reach and has a deadline.
What is stress? Write two techniques to manage it.
Stress is our emotional, mental and physical reaction to a difficult situation. Techniques: time management, physical exercise or yoga, a healthy diet, enough sleep, talking to family and friends, positive thinking. (Any two.)
What is self-motivation? Name its two types.
Self-motivation is the inner force that makes us do things to achieve our goals without being pushed by others. Types: internal motivation (for our own joy and satisfaction) and external motivation (for a reward or praise).
Write two differences between internal and external motivation.
(1) Internal motivation comes from within: we do the work because we enjoy it. External motivation comes from outside: a prize, marks or praise. (2) Internal motivation lasts longer; external motivation ends when the reward stops.
What is self-awareness? Why is it useful?
Self-awareness is knowing your own strengths, weaknesses, values, likes and dislikes. It helps you to choose the right goals, use your strengths and work on your weak points.
Write the four steps of effective time management.
(1) Organise: plan the day. (2) Prioritise: do the most important work first. (3) Control: avoid wasting time. (4) Track: check where the time was spent.
Write any two qualities of a person who can work independently.
Self-awareness, self-motivation, self-regulation, taking responsibility for one's own work and learning continuously. (Any two.)
Rewrite the goal "I want to be fit" as a SMART goal.
"I will jog for 20 minutes every morning, five days a week, and reduce my weight by 3 kg in the next three months." It is specific, measurable, achievable, realistic and has a time limit.
What is self-regulation? Give one example.
Self-regulation is controlling our emotions, thoughts and behaviour to reach long-term goals. Example: switching off the phone while studying even though you want to watch videos.
Write two signs of stress and two effects of stress on studies.
Signs: headache, sleeplessness, sadness, anger. Effects: poor concentration, forgetting what was learnt, low marks. (Any two each.)
Why is goal setting important? Write two points.
(1) It gives a clear direction and keeps us focused. (2) It helps us to plan time, measure progress and stay motivated.
How does a Trojan Horse virus work?
A Trojan Horse looks like a useful program, so the user installs it. Once it runs, it harms the computer: it can steal data and passwords or let other harmful programs in. It does not copy itself like a worm.
What is an operating system? Write two of its functions.
It is the software that works as an interface between the user and the hardware. Functions: manages memory and files, controls devices such as printer and keyboard, runs application programs. (Any two.) Examples: Windows, Linux, Android.
Write two differences between a file and a folder.
(1) A file stores information (text, picture, video); a folder stores files and other folders. (2) A file has an extension such as .txt or .jpg; a folder has no extension.
Write any four ways to protect a computer from threats.
Use strong passwords; install and update antivirus software; turn on the firewall; do not open unknown attachments or links; take regular backups; use only secure (https) sites. (Any four.)
What is a computer virus? Name two types.
A virus is a program made to harm a computer: it can delete data, slow the machine or steal information. Types: worms, Trojan Horse (also boot sector and file viruses).
Write two differences between a worm and a Trojan Horse.
(1) A worm copies itself and spreads from one computer to another on its own; a Trojan Horse does not copy itself. (2) A worm needs no host program; a Trojan Horse hides inside a program that looks useful.
Write any four steps of basic computer maintenance.
Keep the computer clean and dust-free; shut it down properly; take backups; scan for viruses regularly; delete temporary files with a disk cleaner; keep the software updated. (Any four.)
What is a backup? Why is it needed?
A backup is a copy of important data kept on another device or online. If the computer is damaged, stolen or attacked by a virus, the data can be restored from the backup.
What is identity theft? How can you stay safe from it?
A person steals your personal details (name, password, card number) and uses them to cheat. Safety: do not share OTP or passwords, use strong passwords, shop only on secure sites.
Name the file extension for each:
(i) .txt (ii) .jpg or .png (iii) .docx (iv) .mp3
What is a firewall? What does antivirus software do?
A firewall is hardware or software that checks the data coming into and going out of a network and blocks unsafe traffic. Antivirus software finds and removes viruses from the computer.
Write any four qualities of a successful entrepreneur.
Patience, positivity, hard work, never giving up (perseverance), confidence, willingness to try and learn from errors, creativity and innovation. (Any four.)
Who is an entrepreneur? Write two functions of an entrepreneur.
A person who starts and runs a business to meet a need, taking a risk to earn profit. Functions: takes decisions, manages the business, takes risk, brings new ideas or products, divides the income. (Any two.)
Write two ways in which entrepreneurs help society.
(1) They create jobs. (2) They give useful goods and services, often at lower prices, and use local materials. They also add to the income of the country.
Write two differences between wage employment and self-employment.
(1) A wage-employed person works for an employer and gets a fixed salary; a self-employed person works for himself or herself and earns profit. (2) The employer takes the decisions and the risk in wage employment; the self-employed person takes both.
Write any two myths about entrepreneurship and the truth about each.
Myth: every business idea must be unique. Truth: a common idea done in a better way also works. Myth: a lot of money is needed. Truth: many businesses start small. Myth: entrepreneurs are born. Truth: the skills can be learnt.
What are the stages of an entrepreneur's career? Explain in brief.
Enter: start the business. Survive: compete with others and stay in the market. Grow: expand the business to new products or places.
What is innovation? Give one example of an innovative business idea.
Innovation is finding a new or better way of doing something. Example: a service that delivers home-cooked tiffin through a mobile app; edible spoons in place of plastic spoons.
What is meant by risk taking in business? Why is it needed?
It means investing money, time and effort without being sure of the result. Without taking some risk, a new idea can never be tried and the business cannot grow.
Raju sells vegetables on a cart; Meera has started a company that makes paper bags and employs ten people. Who is an entrepreneur? Give a reason.
Both are self-employed, but Meera is an entrepreneur in the full sense: she started with a new idea, took a risk, creates jobs and plans to grow.
Why is entrepreneurship a good career option? Write two points.
(1) You are your own boss and can work on your own idea. (2) There is no limit on earning and growth, and you create jobs for others.
Write two differences between an entrepreneur and an employee.
(1) An entrepreneur owns the business and takes the risk; an employee works in it for a salary. (2) An entrepreneur earns profit, which is not fixed; an employee earns a fixed salary.
What is the role of education in sustainable development?
Education makes people aware of the environment and of their duties. Educated people use resources wisely, adopt clean technology and get better jobs, which reduces poverty. Quality Education is itself SDG 4.
What is sustainable development?
Development that meets the needs of the present generation without harming the ability of future generations to meet their own needs. It balances economic growth, care for the environment and the welfare of society.
What are the Sustainable Development Goals? When were they adopted?
They are 17 goals adopted by the United Nations in 2015 to end poverty, protect the planet and bring peace and prosperity for all by 2030.
Name any four Sustainable Development Goals.
No Poverty (1), Zero Hunger (2), Quality Education (4), Clean Water and Sanitation (6), Affordable and Clean Energy (7), Responsible Consumption and Production (12), Climate Action (13). (Any four.)
Write two things a student can do for sustainable development.
Save water and electricity; say no to single-use plastic and carry a cloth bag; plant trees; use both sides of paper; walk, cycle or use public transport. (Any two.)
What are the 3 Rs? Give one example of each.
Reduce: buy and use less, for example switch off fans when not needed. Reuse: use a thing again, for example a glass bottle. Recycle: make a new product from waste, for example paper from old newspapers.
Which SDG is supported by each of these?
(i) SDG 6: Clean Water and Sanitation. (ii) SDG 7: Affordable and Clean Energy.
Write two problems caused by the over-use of natural resources.
(1) Resources such as coal, petroleum and ground water may finish. (2) Pollution, cutting of forests and climate change harm the health of people and animals.
What is meant by 'Responsible Consumption and Production'? Give one example.
It means producing and using goods in a way that wastes less and harms nature less. Example: using edible or steel cutlery in place of plastic, or buying only what is needed.
What are green skills? Why are they important?
Green skills are the knowledge, abilities and values needed to live and work in a way that protects the environment. They are needed for green jobs such as solar energy, waste management and organic farming.
Write two differences between sustainable and unsustainable development.
(1) Sustainable development saves resources for the future; unsustainable development uses them up quickly. (2) Sustainable development reduces pollution; unsustainable development increases it.
What is bioethics? Write any two of its principles.
Bioethics is a sector-based ethical framework for health care and life sciences. Principles: respect for autonomy, do not harm, ensure maximum benefit for all, give justice. (Any two.)
Name the six stages of the AI Project Cycle in order.
Problem Scoping, Data Acquisition, Data Exploration, Modelling, Evaluation, Deployment.
What is the 4Ws problem canvas?
It is a tool of Problem Scoping. Who: the people who face the problem. What: the nature of the problem. Where: the place or situation in which it occurs. Why: why it should be solved and how the people will benefit.
Name the three domains of AI with one example of each.
Statistical Data: price prediction. Computer Vision: face unlock. Natural Language Processing: voice assistant or language translation.
What is an ethical framework? Why is it needed in AI?
It is a set of values and principles that guides us to take morally right decisions. In AI it is needed so that the system is fair, does not harm people and can be trusted.
Write two differences between sector-based and value-based ethical frameworks.
(1) A sector-based framework is made for one field, such as health care (bioethics); a value-based framework can be used in any field. (2) Value-based frameworks look at the values behind a decision and are of three types: rights-based, utility-based and virtue-based.
Explain rights-based and utility-based frameworks in one line each.
Rights-based: an action is right if it protects the rights and dignity of people. Utility-based: an action is right if it gives the greatest good to the greatest number of people.
What is done in the Data Acquisition stage? Name two sources of data.
Reliable and relevant data is collected for the project. Sources: surveys, sensors, cameras, web scraping, open government portals. (Any two.)
Write the difference between Evaluation and Deployment.
Evaluation: the trained model is tested on new data and its performance is measured. Deployment: the tested model is put to use in the real world for the users.
An AI used for selecting candidates rejects most women because it was trained on old data of male employees. Name the problem and suggest one solution.
The problem is bias in the training data. Solution: train the model on balanced data that represents all groups, and test the results for fairness before use.
What is a virtue-based framework?
It judges an action by the character and intention of the person who acts. It asks whether the action reflects virtues such as honesty, kindness and fairness.
Why is Data Exploration needed before Modelling?
The collected data is large and untidy. Arranging it and drawing graphs shows the trends, patterns and relationships, and helps us to choose the right model.
What is association (association rule learning)? Give an example with products.
It is an unsupervised learning method that finds items which often occur together. Example: customers who buy bread often buy butter also, so a shop keeps them side by side.
What is anomaly detection? Give one example.
It is finding the data points that are very different from the normal pattern. Example: a bank catching an unusually large transaction on a card as a possible fraud.
Write two differences between rule-based and learning-based approaches.
(1) In the rule-based approach the developer gives the rules; in the learning-based approach the machine finds the patterns from data. (2) A rule-based model does not change with new data; a learning-based model improves with more data.
Write two differences between supervised and unsupervised learning.
(1) Supervised learning uses labelled data; unsupervised learning uses unlabelled data. (2) Supervised learning predicts a known output (classification, regression); unsupervised learning finds hidden groups or patterns (clustering, association).
Write two differences between classification and regression.
(1) Classification predicts a category, such as spam or not spam; regression predicts a number, such as price. (2) The output of classification is discrete; the output of regression is continuous.
What is reinforcement learning? Give one example.
The machine learns by trial and error: it gets a reward for a right action and a penalty for a wrong one, and tries to get the maximum reward. Example: an AI learning to play chess, or a robot learning to walk.
What is clustering? Give one example.
Clustering is unsupervised learning that puts similar data points into groups (clusters) without any labels. Example: grouping the customers of a mall by their buying habits.
Explain the relation between AI, ML and DL.
AI is the broad field of making machines intelligent. Machine Learning is a part of AI in which machines learn from data. Deep Learning is a part of ML that uses neural networks with many layers.
What are features and labels? Give an example.
Features are the input columns used for prediction; the label is the output to be predicted. Example: to predict the price of a house, area and number of rooms are features and price is the label.
What is the difference between training data and testing data?
Training data is used to teach the model. Testing data is kept aside and is used to check how well the trained model works on data it has not seen.
Name the three kinds of layers of an artificial neural network and the work of each.
Input layer: receives the data. Hidden layers: process the data using weights and bias. Output layer: gives the final result.
What are weights and bias in a neural network?
A weight is a number that shows how important an input is. The bias is an extra number added to the weighted sum. Both are adjusted during training so that the output becomes correct.
Identify the type of learning:
(i) Supervised learning (regression). (ii) Unsupervised learning (clustering).
Identify the model:
(i) Classification (supervised). (ii) Association (unsupervised).
What is an artificial neural network? How is it like the human brain?
It is a learning model made of layers of connected units called neurons. Like the neurons of the brain, each unit receives signals, processes them and passes the result to the next units.
Write one advantage and one limitation of a rule-based model.
Advantage: it is simple, and its decisions are easy to explain. Limitation: it cannot handle situations for which no rule was written, and it does not learn from new data.
What is labelled data? Give one example of labelled and unlabelled data.
Labelled data has the correct answer attached to every record. Labelled: photos tagged 'cat' or 'dog'. Unlabelled: a folder of photos with no tags.
Which type of learning is suitable for each?
(i) Reinforcement learning. (ii) Supervised learning (classification).
What is deep learning? Name two of its uses.
Deep learning is machine learning with neural networks that have many hidden layers; the machine finds the features itself. Uses: face recognition, speech recognition, self-driving cars. (Any two.)
What is an activation function?
It is the function applied to the weighted sum of a neuron to decide its output, that is, whether and how strongly the neuron passes the signal to the next layer.
Why does a learning-based model need a large amount of data?
The model finds the rules from examples. More and varied examples help it to learn the true pattern, so it makes fewer mistakes on new data.
Name any two sub-categories of unsupervised learning and write one use of each.
Clustering: grouping customers. Association: finding products bought together. (Anomaly detection: catching fraud.)
Why is model evaluation needed? Why is the model tested on unseen data?
Evaluation tells how well the model works and whether it can be trusted before deployment. Testing on unseen data shows whether the model has really learnt the pattern; testing on training data only shows what it memorised.
What is a train-test split?
The data set is divided into two parts: a bigger training set to teach the model (for example 80%) and a smaller testing set (20%) to check it on unseen data.
Define True Positive and False Negative with an example of a disease test.
True Positive: the model says 'sick' and the person is sick. False Negative: the model says 'healthy' but the person is sick.
Write the formulas of precision and recall.
Precision = TP / (TP + FP). Recall = TP / (TP + FN).
What is the F1 score? Write its formula.
The F1 score combines precision and recall in one number (their harmonic mean). F1 = 2 x Precision x Recall / (Precision + Recall). It is high only when both are high.
Which metric is more important for a model that detects forest fire, precision or recall? Why?
Recall. Missing a real fire (False Negative) is very dangerous, so the model must catch every fire even if it gives a few false alarms.
Which metric is more important for a spam filter, precision or recall? Why?
Precision. If a genuine mail is marked as spam (False Positive), the user may lose an important message.
Why can accuracy be misleading? Give an example.
When one class is very rare, a model can score high accuracy by always predicting the common class. Example: if 2 of 100 people are sick, a model saying 'healthy' for all is 98% accurate but finds no patient.
What is overfitting?
The model learns the training data too closely, including its noise. It gives very good results on training data but poor results on new data.
Write two ethical concerns in model evaluation.
Bias: the model may work well for one group and badly for another. Transparency: people should be able to know how the model was tested and how it decides. Accuracy: results should be reported honestly. (Any two.)
What is a confusion matrix?
It is a table that compares the predictions of a model with the actual values. For two classes it has four cells: True Positive, True Negative, False Positive and False Negative.
Write the difference between object detection and instance segmentation.
Object detection finds the objects in an image and marks each with a box. Instance segmentation goes further: it marks the exact pixels (outline) of every separate object.
What is Computer Vision? Name two applications.
Computer Vision is the AI domain that lets machines see, process and understand images and videos. Applications: face recognition, self-driving cars, medical imaging, search by image. (Any two.)
What is a pixel? What is resolution?
A pixel (picture element) is the smallest unit of a digital image. Resolution is the number of pixels in the image, written as width x height, for example 1280 x 720.
Write two differences between a grayscale image and an RGB image.
(1) A grayscale image has one channel with shades from black (0) to white (255); an RGB image has three channels: red, green and blue. (2) A grayscale pixel stores one value (1 byte); an RGB pixel stores three values (3 bytes).
Name the four tasks of Computer Vision.
For a single object: image classification, classification + localisation. For multiple objects: object detection, instance segmentation.
How does a computer store an image?
As a grid of numbers. Each pixel has a value from 0 to 255 (one value in grayscale, three values for red, green and blue in a colour image).
What is image classification? Give one example.
Giving one label to the whole image out of a fixed set of classes. Example: telling whether an X-ray is 'normal' or 'not normal'.
What colour do these RGB pixels show?
(i) Black (ii) Green (iii) Yellow (red + green) (iv) White.
Write two uses of Computer Vision in health care or agriculture.
Health care: finding tumours in X-ray and MRI images. Agriculture: finding diseased leaves or weeds from photos taken by a drone.
What is classification + localisation?
The model tells what the single object in the image is (classification) and also where it is, by drawing a box around it (localisation).
What is a convolution? What is it used for?
Convolution multiplies a small matrix called a kernel with parts of the image, one after another, to make a new image. It brings out features such as edges and is the main step of a CNN.
Name the layers of a Convolutional Neural Network.
Convolution layer, Rectified Linear Unit (ReLU) layer, pooling layer and fully connected layer.
A news website sorts its articles into sports, business and entertainment automatically. Name the NLP application and write one benefit.
Text classification. Benefit: thousands of articles are sorted in seconds without manual work, so readers find the news of their interest quickly.
What is Natural Language Processing? Name two applications.
NLP is the AI domain that lets machines understand, interpret and produce human language. Applications: voice assistants, language translation, sentiment analysis, automatic captions, text classification. (Any two.)
Write two differences between a script bot and a smart bot.
(1) A script bot works on a fixed script of rules; a smart bot learns from data using AI. (2) A script bot is easy to make but has limited use; a smart bot is flexible and improves with use (example: Alexa).
Write two differences between stemming and lemmatisation.
(1) Stemming cuts the affixes and the result may not be a meaningful word (studies gives studi); lemmatisation gives a meaningful word (studies gives study). (2) Stemming is faster; lemmatisation takes more time.
What are stop words? Give four examples.
Stop words are words that occur very often in a text but add little meaning. Examples: a, an, the, is, are, to, and, of. They are usually removed in text normalisation.
Name the steps of text normalisation in order.
Sentence segmentation, tokenisation, removal of stop words, special characters and numbers, converting to a common case, stemming, lemmatisation.
What is the Bag of Words model?
It is an NLP model that takes out the unique words of the corpus (the vocabulary) and counts how many times each word occurs in each document. The order of words is not kept.
What is sentiment analysis? Where is it used?
Finding the feeling (positive, negative or neutral) expressed in a text. Companies use it on product reviews and social media posts to know what customers think.
What is the difference between syntax and semantics?
Syntax is the grammar: the correct arrangement of words in a sentence. Semantics is the meaning of the words and the sentence. A sentence can have correct syntax and still no meaning.
Why is human language difficult for a computer to understand? Write two reasons.
(1) The same word can have many meanings (bank of a river, money in the bank). (2) A perfectly grammatical sentence may have no meaning, and meaning also depends on the situation.
What is TF-IDF? Why is it better than a simple word count?
TF-IDF = Term Frequency x Inverse Document Frequency. It gives a high value to words that are frequent in one document but rare in the corpus, so the important words stand out and common words get a low value.
What is a corpus? What is a token?
A corpus is the whole collection of text of a project. A token is one unit of a sentence: a word, a number or a special character.
A model was tested on 100 cases. The results are: TP = 48, TN = 34, FP = 16, FN = 2. Calculate the accuracy and the error rate of the model.
Accuracy = (TP + TN) / Total = (48 + 34) / 100 = 82 / 100 = 0.82 (82%)
Error rate = 1 - Accuracy = (FP + FN) / Total = 18 / 100 = 0.18 (18%)
For a model, TP = 12, FP = 13 and FN = 12. Calculate its precision and recall.
Precision = TP / (TP + FP) = 12 / (12 + 13) = 12 / 25 = 0.48 (48%)
Recall = TP / (TP + FN) = 12 / (12 + 12) = 12 / 24 = 0.50 (50%)
The precision of a model is 0.90 and its recall is 0.60. Calculate the F1 score.
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.90 x 0.60 / (0.90 + 0.60) = 1.08 / 1.50 = 0.72
A model was tested on 200 cases. The results are: TP = 36, TN = 104, FP = 24, FN = 36. Calculate the accuracy and the error rate of the model.
Accuracy = (TP + TN) / Total = (36 + 104) / 200 = 140 / 200 = 0.70 (70%)
Error rate = 1 - Accuracy = (FP + FN) / Total = 60 / 200 = 0.30 (30%)
For a model, TP = 72, FP = 8 and FN = 28. Calculate its precision and recall.
Precision = TP / (TP + FP) = 72 / (72 + 8) = 72 / 80 = 0.90 (90%)
Recall = TP / (TP + FN) = 72 / (72 + 28) = 72 / 100 = 0.72 (72%)
The precision of a model is 0.50 and its recall is 0.60. Calculate the F1 score.
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.50 x 0.60 / (0.50 + 0.60) = 0.6 / 1.10 = 0.55
A model was tested on 50 cases. The results are: TP = 21, TN = 13, FP = 9, FN = 7. Calculate the accuracy and the error rate of the model.
Accuracy = (TP + TN) / Total = (21 + 13) / 50 = 34 / 50 = 0.68 (68%)
Error rate = 1 - Accuracy = (FP + FN) / Total = 16 / 50 = 0.32 (32%)
For a model, TP = 9, FP = 6 and FN = 3. Calculate its precision and recall.
Precision = TP / (TP + FP) = 9 / (9 + 6) = 9 / 15 = 0.60 (60%)
Recall = TP / (TP + FN) = 9 / (9 + 3) = 9 / 12 = 0.75 (75%)
The precision of a model is 0.90 and its recall is 0.60. Calculate the F1 score.
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.90 x 0.60 / (0.90 + 0.60) = 1.08 / 1.50 = 0.72
A model was tested on 100 cases. The results are: TP = 36, TN = 28, FP = 24, FN = 12. Calculate the accuracy and the error rate of the model.
Accuracy = (TP + TN) / Total = (36 + 28) / 100 = 64 / 100 = 0.64 (64%)
Error rate = 1 - Accuracy = (FP + FN) / Total = 36 / 100 = 0.36 (36%)
For a model, TP = 24, FP = 16 and FN = 8. Calculate its precision and recall.
Precision = TP / (TP + FP) = 24 / (24 + 16) = 24 / 40 = 0.60 (60%)
Recall = TP / (TP + FN) = 24 / (24 + 8) = 24 / 32 = 0.75 (75%)
The precision of a model is 0.60 and its recall is 0.40. Calculate the F1 score.
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.60 x 0.40 / (0.60 + 0.40) = 0.48 / 1.00 = 0.48
A model predicted the price of a shirt (in rupees) as 270. The actual value is 250. Calculate the error, the error rate and the accuracy of the prediction.
Error = |actual - predicted| = |250 - 270| = 20. Error rate = error / actual = 20 / 250 = 0.08 = 8%. Accuracy = 100% - 8% = 92%
A model predicted the marks of a class (total) as 380. The actual value is 400. Calculate the error, the error rate and the accuracy of the prediction.
Error = |actual - predicted| = |400 - 380| = 20. Error rate = error / actual = 20 / 400 = 0.05 = 5%. Accuracy = 100% - 5% = 95%
A model predicted the daily visitors of a park as 560. The actual value is 500. Calculate the error, the error rate and the accuracy of the prediction.
Error = |actual - predicted| = |500 - 560| = 60. Error rate = error / actual = 60 / 500 = 0.12 = 12%. Accuracy = 100% - 12% = 88%
A model predicted the units of electricity used as 720. The actual value is 800. Calculate the error, the error rate and the accuracy of the prediction.
Error = |actual - predicted| = |800 - 720| = 80. Error rate = error / actual = 80 / 800 = 0.10 = 10%. Accuracy = 100% - 10% = 90%
A model predicted the weight of a bag of rice (in grams) as 1040. The actual value is 1000. Calculate the error, the error rate and the accuracy of the prediction.
Error = |actual - predicted| = |1000 - 1040| = 40. Error rate = error / actual = 40 / 1000 = 0.04 = 4%. Accuracy = 100% - 4% = 96%
A model predicted the monthly sale of a shop (in rupees) as 1700. The actual value is 2000. Calculate the error, the error rate and the accuracy of the prediction.
Error = |actual - predicted| = |2000 - 1700| = 300. Error rate = error / actual = 300 / 2000 = 0.15 = 15%. Accuracy = 100% - 15% = 85%
Perform these steps of text normalisation on the sentence "Raj and Vicky are best friends.":
7 tokens: Raj | and | Vicky | are | best | friends | .
4 words left: raj | vicky | best | friends
Two documents are given. Document 1: "Trees give us oxygen." Document 2: "We must plant trees." Write the vocabulary and the document vector of each document (change to lower case, remove punctuation, keep all the words).
Vocabulary: trees, give, us, oxygen, we, must, plant
| trees | give | us | oxygen | we | must | plant | |
|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 |
| Document 2 | 1 | 0 | 0 | 0 | 1 | 1 | 1 |
Perform these steps of text normalisation on the sentence "We are learning Python at school.":
7 tokens: We | are | learning | Python | at | school | .
3 words left: learning | python | school
Two documents are given. Document 1: "We love our school." Document 2: "Our school has a big library." Write the vocabulary and the document vector of each document (change to lower case, remove punctuation, keep all the words).
Vocabulary: we, love, our, school, has, a, big, library
| we | love | our | school | has | a | big | library | |
|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 1 | 1 | 1 | 1 | 1 | 1 |
Perform these steps of text normalisation on the sentence "Is it raining today?":
5 tokens: Is | it | raining | today | ?
2 words left: raining | today
Two documents are given. Document 1: "Honey bees make honey." Document 2: "Bees live in a hive." Write the vocabulary and the document vector of each document (change to lower case, remove punctuation, keep all the words).
Vocabulary: honey, bees, make, live, in, a, hive
| honey | bees | make | live | in | a | hive | |
|---|---|---|---|---|---|---|---|
| Document 1 | 2 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 1 | 0 | 1 | 1 | 1 | 1 |
Perform these steps of text normalisation on the sentence "AI is changing the world, right?":
8 tokens: AI | is | changing | the | world | , | right | ?
4 words left: ai | changing | world | right
Two documents are given. Document 1: "Reading is a good habit." Document 2: "A good habit makes a good life." Write the vocabulary and the document vector of each document (change to lower case, remove punctuation, keep all the words).
Vocabulary: reading, is, a, good, habit, makes, life
| reading | is | a | good | habit | makes | life | |
|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 |
| Document 2 | 0 | 0 | 2 | 2 | 1 | 1 | 1 |
Perform these steps of text normalisation on the sentence "The cat sat on the mat.":
7 tokens: The | cat | sat | on | the | mat | .
3 words left: cat | sat | mat
Two documents are given. Document 1: "Rain helps the farmers." Document 2: "Farmers grow rice in rain." Write the vocabulary and the document vector of each document (change to lower case, remove punctuation, keep all the words).
Vocabulary: rain, helps, the, farmers, grow, rice, in
| rain | helps | the | farmers | grow | rice | in | |
|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 |
| Document 2 | 1 | 0 | 0 | 1 | 1 | 1 | 1 |
Perform these steps of text normalisation on the sentence "Wow! We won the match.":
7 tokens: Wow | ! | We | won | the | match | .
3 words left: wow | won | match
Two documents are given. Document 1: "Chess is a mind game." Document 2: "Asha plays chess well." Write the vocabulary and the document vector of each document (change to lower case, remove punctuation, keep all the words).
Vocabulary: chess, is, a, mind, game, asha, plays, well
| chess | is | a | mind | game | asha | plays | well | |
|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 |
| Document 2 | 1 | 0 | 0 | 0 | 0 | 1 | 1 | 1 |
An image has the resolution 640 x 480.
Pixels = 640 x 480 = 307,200
Grayscale: 1 byte per pixel = 307,200 bytes. RGB: 3 bytes per pixel = 307,200 x 3 = 921,600 bytes
An image has the resolution 1280 x 720.
Pixels = 1280 x 720 = 921,600
Grayscale: 1 byte per pixel = 921,600 bytes. RGB: 3 bytes per pixel = 921,600 x 3 = 2,764,800 bytes
An image has the resolution 100 x 100.
Pixels = 100 x 100 = 10,000
Grayscale: 1 byte per pixel = 10,000 bytes. RGB: 3 bytes per pixel = 10,000 x 3 = 30,000 bytes
An image has the resolution 1920 x 1080.
Pixels = 1920 x 1080 = 2,073,600
Grayscale: 1 byte per pixel = 2,073,600 bytes. RGB: 3 bytes per pixel = 2,073,600 x 3 = 6,220,800 bytes
An image has the resolution 800 x 600.
Pixels = 800 x 600 = 480,000
Grayscale: 1 byte per pixel = 480,000 bytes. RGB: 3 bytes per pixel = 480,000 x 3 = 1,440,000 bytes
An image has the resolution 256 x 256.
Pixels = 256 x 256 = 65,536
Grayscale: 1 byte per pixel = 65,536 bytes. RGB: 3 bytes per pixel = 65,536 x 3 = 196,608 bytes
A city uses AI cameras with face recognition in all public places to catch criminals. Discuss two ethical concerns. Suggest two safeguards that protect the privacy of citizens.
Concerns: (1) Privacy: every person is watched and recorded without consent. (2) Bias and wrong matches: the system may wrongly identify innocent people, more often from some groups. Data may also be misused or leaked. Safeguards: (1) Clear rules on what is recorded, who can see it and for how long it is kept. (2) Regular checks of the system for bias and accuracy, with a human taking the final decision; strong data security and a way for citizens to complain.
Explain the six stages of the AI Project Cycle with one line on each.
Problem Scoping: understand the problem using the 4Ws. Data Acquisition: collect reliable data. Data Exploration: arrange the data and draw graphs to find patterns. Modelling: choose and train a model. Evaluation: test the model on new data. Deployment: put the model to use for real users.
A school wants an AI system that warns teachers about students who may fail. Fill the 4Ws problem canvas for it.
Who: students at risk of failing, their teachers and parents. What: weak students are identified too late, after the result. Where: in the school, during the session, in class tests and attendance. Why: early help will improve learning and results, and reduce stress.
Explain the four principles of bioethics with an example of an AI tool that suggests treatment.
Respect for autonomy: the patient and doctor must be told how the tool works and may refuse its advice. Do not harm: the tool must not suggest a treatment that injures the patient. Ensure maximum benefit for all: it should give the best possible result to every patient. Give justice: it must work equally well for all groups, with no bias of gender, region or income.
What are value-based frameworks? Explain the three types with one line each.
Value-based frameworks look at the moral values that guide a decision. Rights-based: protect human rights and dignity above everything. Utility-based: choose the action that gives the greatest good to the greatest number. Virtue-based: act as a person of good character would: honestly, kindly and fairly.
A hospital's AI for skin disease was trained mainly on images of one skin type and makes more mistakes for other patients. (a) Name the ethical problem. (b) Which principle of bioethics is broken? (c) Suggest two corrections.
(a) Bias in the training data. (b) 'Give justice' (and 'do not harm'): all patients are not served equally. (c) Collect training images of all skin types; test the model separately for each group before use; keep a doctor to check the result.
Name the three domains of AI. For each, write what kind of data it uses and one application.
Statistical Data (Data Science): numbers and tables; price prediction, weather forecast. Computer Vision: images and videos; face unlock, self-driving cars. Natural Language Processing: text and speech; voice assistants, translation.
A bank uses AI to approve loans. Applicants are not told why a loan is rejected. Discuss the issue using an ethical framework and suggest improvements.
The issue is lack of transparency and possible unfairness. A rights-based view says every applicant has a right to fair treatment and to an explanation. Improvements: give the reason for each decision; check the data and results for bias; allow an appeal to a human officer; protect the applicants' data.
Why do we need ethical frameworks for AI? Explain with two examples.
AI systems take decisions that affect people's lives, and they can cause harm without intending to. Frameworks give a structured way to think before building. Examples: (1) a hiring AI trained on old data may reject women; a framework makes us test for bias. (2) a health app may share patient data; a framework makes us protect privacy and take consent.
Differentiate between Data Acquisition and Data Exploration. Why are both needed before Modelling?
Data Acquisition collects relevant, reliable data from sources such as surveys, sensors and websites. Data Exploration arranges that data and shows it as graphs to find patterns and errors. A model learns only from data: wrong or poorly understood data gives a wrong model.
A self-driving car must choose between two risky actions in an accident. How would a utility-based and a rights-based framework look at the choice?
Utility-based: choose the action that saves the most lives or causes the least total harm. Rights-based: every person has an equal right to life and safety, so no person should be treated only as a means to save others. The two can give different answers, which is why clear public rules are needed.
Explain the layers of an artificial neural network. How do they help in recognising an object in an image?
Input layer: receives the pixel values of the image. Hidden layers: each neuron multiplies its inputs by weights, adds a bias and applies an activation function; the early layers find edges and simple shapes, the later layers combine them into parts and objects. Output layer: gives the final answer, for example 'cat' or 'dog'. During training the weights and biases are adjusted until the answers are correct.
A company wants a chatbot for customer support. Should it choose the rule-based or the learning-based approach? Compare the two and justify.
Rule-based: replies come from fixed 'if-then' rules; cheap and predictable, good for a few standard questions (order status, timings), but it fails on unexpected questions. Learning-based: the bot learns from past chats; it understands different ways of asking and improves with use, but needs a lot of data and testing. For a small set of fixed questions choose rule-based; for varied questions from many customers choose learning-based.
Explain supervised, unsupervised and reinforcement learning with one example of each.
Supervised: learns from labelled data; example: spam detection from marked e-mails. Unsupervised: finds hidden groups in unlabelled data; example: grouping customers by buying habits. Reinforcement: learns by trial and error using rewards and penalties; example: an AI learning to play a game.
Explain classification, regression, clustering and association with one example each.
Classification (supervised): predicts a category; mail is spam or not spam. Regression (supervised): predicts a number; price of a house. Clustering (unsupervised): groups similar items; types of customers. Association (unsupervised): finds items that occur together; bread and butter.
Compare the rule-based and learning-based approaches in four points.
(1) Rules are written by the developer / patterns are learnt from data. (2) Needs little data / needs a lot of data. (3) Does not change after it is made / improves with new data. (4) Fails in new situations / can handle new situations. Example: a fixed-reply bot / a spam filter that learns.
Explain how one neuron of a neural network takes a decision. Use the terms input, weight, bias and activation function.
Each input is multiplied by its weight, which shows its importance. The products are added and the bias is added to the sum. The activation function is applied to this weighted sum; for a simple step function the neuron gives 1 if the sum reaches the threshold and 0 otherwise. The output goes to the next layer.
Identify the type of learning and its sub-category: (a) predicting rainfall in mm, (b) sorting photos into 'cat' and 'dog' using tagged photos, (c) a shop finding groups of similar customers, (d) a robot learning to walk by trial and error.
(a) Supervised, regression. (b) Supervised, classification. (c) Unsupervised, clustering. (d) Reinforcement learning.
Explain AI, Machine Learning and Deep Learning. How are they related? Give one example of each.
AI: any technique that lets a machine act intelligently; a rule-based chess program. ML: a part of AI in which the machine learns from data; a spam filter. DL: a part of ML that uses neural networks with many layers; face recognition. DL is inside ML, and ML is inside AI.
An online store wants to (a) predict next month's sales, (b) group its customers, (c) find products bought together, (d) catch fake orders. Name the technique for each.
(a) Regression (supervised). (b) Clustering (unsupervised). (c) Association (unsupervised). (d) Anomaly detection.
What are the common terms used for data in AI? Explain features, labels, training data and testing data with an example of predicting whether a student will pass.
Features: the inputs, such as attendance and study hours. Label: the output to predict, 'pass' or 'fail'. Training data: records with features and labels used to teach the model. Testing data: records kept aside to check the trained model on unseen cases.
Write two advantages and two limitations of neural networks.
Advantages: (1) they find the features on their own, so no rules need to be written; (2) they work very well on images, speech and text. Limitations: (1) they need a large amount of data and computing power; (2) their decisions are hard to explain (a 'black box').
Explain the steps of text normalisation with the sentence: "The children are playing happily."
Sentence segmentation: one sentence. Tokenisation: The | children | are | playing | happily | . Removing stop words and special characters: children | playing | happily. Common case: all in lower case. Stemming: children, play, happili. Lemmatisation: child, play, happily (meaningful words).
Write the four steps of the Bag of Words algorithm.
(1) Text normalisation: collect the data and pre-process it. (2) Create the dictionary: list all the unique words of the corpus. (3) Create a document vector: for one document, write how many times each word of the dictionary occurs. (4) Repeat for all the documents to get the document vector table.
Explain the four tasks of Computer Vision with one example of each.
Image classification: one label for the image; cat or dog. Classification + localisation: the label and a box around the single object. Object detection: boxes and labels for many objects; cars and people on a road. Instance segmentation: the exact outline of each object; every person in a crowd marked separately.
Explain precision, recall and F1 score. When is each one important? Give an example for precision and for recall.
Precision = TP / (TP + FP): of the cases predicted positive, how many are right. Important when a False Positive is costly, such as a spam filter. Recall = TP / (TP + FN): of the real positives, how many were found. Important when a False Negative is costly, such as a disease test. F1 = 2PR / (P + R) balances the two and is used when both matter or the classes are unequal.
The confusion matrix of a disease-screening AI, tested on 50 cases, is given below. Calculate:
Confusion matrix
| Reality: has the disease | Reality: is healthy | |
|---|---|---|
| Predicted: has the disease | 21TP | 14FP |
| Predicted: is healthy | 7FN | 8TN |
Accuracy = (TP + TN) / Total = (21 + 8) / 50 = 29 / 50 = 0.58 (58%)
Precision = TP / (TP + FP) = 21 / (21 + 14) = 21 / 35 = 0.60 (60%)
Recall = TP / (TP + FN) = 21 / (21 + 7) = 21 / 28 = 0.75 (75%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.60 x 0.75 / (0.60 + 0.75) = 0.9 / 1.35 = 0.67
A spam filter was tested on 50 cases. It predicted 'spam' for 30 cases; out of these, 18 were really 'spam'. It predicted 'genuine mail' for the other 20 cases; out of these, 13 were really 'genuine mail'.
TP = 18, FP = 30 - 18 = 12, TN = 13, FN = 20 - 13 = 7.
Confusion matrix
| Reality: spam | Reality: genuine mail | |
|---|---|---|
| Predicted: spam | 18TP | 12FP |
| Predicted: genuine mail | 7FN | 13TN |
Accuracy = (TP + TN) / Total = (18 + 13) / 50 = 31 / 50 = 0.62 (62%)
Precision = TP / (TP + FP) = 18 / (18 + 12) = 18 / 30 = 0.60 (60%)
Recall = TP / (TP + FN) = 18 / (18 + 7) = 18 / 25 = 0.72 (72%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.60 x 0.72 / (0.60 + 0.72) = 0.864 / 1.32 = 0.65
Precision, because a genuine mail sent to the spam folder (False Positive) may be an important message.
The confusion matrix of a forest-fire alert system, tested on 1000 cases, is given below. Calculate:
Confusion matrix
| Reality: fire | Reality: no fire | |
|---|---|---|
| Predicted: fire | 240TP | 80FP |
| Predicted: no fire | 260FN | 420TN |
Accuracy = (TP + TN) / Total = (240 + 420) / 1000 = 660 / 1000 = 0.66 (66%)
Precision = TP / (TP + FP) = 240 / (240 + 80) = 240 / 320 = 0.75 (75%)
Recall = TP / (TP + FN) = 240 / (240 + 260) = 240 / 500 = 0.48 (48%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.75 x 0.48 / (0.75 + 0.48) = 0.72 / 1.23 = 0.59
A bank's fraud detector was tested on 1000 cases. It predicted 'fraud' for 520 cases; out of these, 260 were really 'fraud'. It predicted 'genuine transaction' for the other 480 cases; out of these, 240 were really 'genuine transaction'.
TP = 260, FP = 520 - 260 = 260, TN = 240, FN = 480 - 240 = 240.
Confusion matrix
| Reality: fraud | Reality: genuine transaction | |
|---|---|---|
| Predicted: fraud | 260TP | 260FP |
| Predicted: genuine transaction | 240FN | 240TN |
Accuracy = (TP + TN) / Total = (260 + 240) / 1000 = 500 / 1000 = 0.50 (50%)
Precision = TP / (TP + FP) = 260 / (260 + 260) = 260 / 520 = 0.50 (50%)
Recall = TP / (TP + FN) = 260 / (260 + 240) = 260 / 500 = 0.52 (52%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.50 x 0.52 / (0.50 + 0.52) = 0.52 / 1.02 = 0.51
Recall, because a fraud that is missed (False Negative) causes a loss of money.
The confusion matrix of a plagiarism checker, tested on 200 cases, is given below. Calculate:
Confusion matrix
| Reality: copied work | Reality: original work | |
|---|---|---|
| Predicted: copied work | 32TP | 48FP |
| Predicted: original work | 8FN | 112TN |
Accuracy = (TP + TN) / Total = (32 + 112) / 200 = 144 / 200 = 0.72 (72%)
Precision = TP / (TP + FP) = 32 / (32 + 48) = 32 / 80 = 0.40 (40%)
Recall = TP / (TP + FN) = 32 / (32 + 8) = 32 / 40 = 0.80 (80%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.40 x 0.80 / (0.40 + 0.80) = 0.64 / 1.20 = 0.53
A face-unlock system (positive = owner) was tested on 500 cases. It predicted 'owner' for 250 cases; out of these, 150 were really 'owner'. It predicted 'stranger' for the other 250 cases; out of these, 200 were really 'stranger'.
TP = 150, FP = 250 - 150 = 100, TN = 200, FN = 250 - 200 = 50.
Confusion matrix
| Reality: owner | Reality: stranger | |
|---|---|---|
| Predicted: owner | 150TP | 100FP |
| Predicted: stranger | 50FN | 200TN |
Accuracy = (TP + TN) / Total = (150 + 200) / 500 = 350 / 500 = 0.70 (70%)
Precision = TP / (TP + FP) = 150 / (150 + 100) = 150 / 250 = 0.60 (60%)
Recall = TP / (TP + FN) = 150 / (150 + 50) = 150 / 200 = 0.75 (75%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.60 x 0.75 / (0.60 + 0.75) = 0.9 / 1.35 = 0.67
Precision, because a stranger accepted as the owner (False Positive) can open the phone.
The confusion matrix of a flood-warning model, tested on 1000 cases, is given below. Calculate:
Confusion matrix
| Reality: flood | Reality: no flood | |
|---|---|---|
| Predicted: flood | 260TP | 140FP |
| Predicted: no flood | 260FN | 340TN |
Accuracy = (TP + TN) / Total = (260 + 340) / 1000 = 600 / 1000 = 0.60 (60%)
Precision = TP / (TP + FP) = 260 / (260 + 140) = 260 / 400 = 0.65 (65%)
Recall = TP / (TP + FN) = 260 / (260 + 260) = 260 / 520 = 0.50 (50%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.65 x 0.50 / (0.65 + 0.50) = 0.65 / 1.15 = 0.57
An AI that finds pneumonia in chest X-rays was tested on 50 cases. It predicted 'pneumonia' for 25 cases; out of these, 24 were really 'pneumonia'. It predicted 'normal' for the other 25 cases; out of these, 19 were really 'normal'.
TP = 24, FP = 25 - 24 = 1, TN = 19, FN = 25 - 19 = 6.
Confusion matrix
| Reality: pneumonia | Reality: normal | |
|---|---|---|
| Predicted: pneumonia | 24TP | 1FP |
| Predicted: normal | 6FN | 19TN |
Accuracy = (TP + TN) / Total = (24 + 19) / 50 = 43 / 50 = 0.86 (86%)
Precision = TP / (TP + FP) = 24 / (24 + 1) = 24 / 25 = 0.96 (96%)
Recall = TP / (TP + FN) = 24 / (24 + 6) = 24 / 30 = 0.80 (80%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.96 x 0.80 / (0.96 + 0.80) = 1.536 / 1.76 = 0.87
Recall, because a patient whose pneumonia is missed (False Negative) goes untreated.
The confusion matrix of a filter that blocks unsafe videos for children, tested on 100 cases, is given below. Calculate:
Confusion matrix
| Reality: unsafe video | Reality: safe video | |
|---|---|---|
| Predicted: unsafe video | 24TP | 6FP |
| Predicted: safe video | 16FN | 54TN |
Accuracy = (TP + TN) / Total = (24 + 54) / 100 = 78 / 100 = 0.78 (78%)
Precision = TP / (TP + FP) = 24 / (24 + 6) = 24 / 30 = 0.80 (80%)
Recall = TP / (TP + FN) = 24 / (24 + 16) = 24 / 40 = 0.60 (60%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.80 x 0.60 / (0.80 + 0.60) = 0.96 / 1.40 = 0.69
A camera that finds defective bottles in a factory was tested on 200 cases. It predicted 'defective bottle' for 120 cases; out of these, 84 were really 'defective bottle'. It predicted 'good bottle' for the other 80 cases; out of these, 24 were really 'good bottle'.
TP = 84, FP = 120 - 84 = 36, TN = 24, FN = 80 - 24 = 56.
Confusion matrix
| Reality: defective bottle | Reality: good bottle | |
|---|---|---|
| Predicted: defective bottle | 84TP | 36FP |
| Predicted: good bottle | 56FN | 24TN |
Accuracy = (TP + TN) / Total = (84 + 24) / 200 = 108 / 200 = 0.54 (54%)
Precision = TP / (TP + FP) = 84 / (84 + 36) = 84 / 120 = 0.70 (70%)
Recall = TP / (TP + FN) = 84 / (84 + 56) = 84 / 140 = 0.60 (60%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.70 x 0.60 / (0.70 + 0.60) = 0.84 / 1.30 = 0.65
Recall, because a defective bottle that is passed (False Negative) reaches the customer.
The confusion matrix of a disease-screening AI, tested on 1000 cases, is given below. Calculate:
Confusion matrix
| Reality: has the disease | Reality: is healthy | |
|---|---|---|
| Predicted: has the disease | 120TP | 40FP |
| Predicted: is healthy | 180FN | 660TN |
Accuracy = (TP + TN) / Total = (120 + 660) / 1000 = 780 / 1000 = 0.78 (78%)
Precision = TP / (TP + FP) = 120 / (120 + 40) = 120 / 160 = 0.75 (75%)
Recall = TP / (TP + FN) = 120 / (120 + 180) = 120 / 300 = 0.40 (40%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.75 x 0.40 / (0.75 + 0.40) = 0.6 / 1.15 = 0.52
A spam filter was tested on 200 cases. It predicted 'spam' for 112 cases; out of these, 84 were really 'spam'. It predicted 'genuine mail' for the other 88 cases; out of these, 32 were really 'genuine mail'.
TP = 84, FP = 112 - 84 = 28, TN = 32, FN = 88 - 32 = 56.
Confusion matrix
| Reality: spam | Reality: genuine mail | |
|---|---|---|
| Predicted: spam | 84TP | 28FP |
| Predicted: genuine mail | 56FN | 32TN |
Accuracy = (TP + TN) / Total = (84 + 32) / 200 = 116 / 200 = 0.58 (58%)
Precision = TP / (TP + FP) = 84 / (84 + 28) = 84 / 112 = 0.75 (75%)
Recall = TP / (TP + FN) = 84 / (84 + 56) = 84 / 140 = 0.60 (60%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.75 x 0.60 / (0.75 + 0.60) = 0.9 / 1.35 = 0.67
Precision, because a genuine mail sent to the spam folder (False Positive) may be an important message.
The confusion matrix of a forest-fire alert system, tested on 100 cases, is given below. Calculate:
Confusion matrix
| Reality: fire | Reality: no fire | |
|---|---|---|
| Predicted: fire | 36TP | 24FP |
| Predicted: no fire | 14FN | 26TN |
Accuracy = (TP + TN) / Total = (36 + 26) / 100 = 62 / 100 = 0.62 (62%)
Precision = TP / (TP + FP) = 36 / (36 + 24) = 36 / 60 = 0.60 (60%)
Recall = TP / (TP + FN) = 36 / (36 + 14) = 36 / 50 = 0.72 (72%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.60 x 0.72 / (0.60 + 0.72) = 0.864 / 1.32 = 0.65
A bank's fraud detector was tested on 100 cases. It predicted 'fraud' for 30 cases; out of these, 24 were really 'fraud'. It predicted 'genuine transaction' for the other 70 cases; out of these, 54 were really 'genuine transaction'.
TP = 24, FP = 30 - 24 = 6, TN = 54, FN = 70 - 54 = 16.
Confusion matrix
| Reality: fraud | Reality: genuine transaction | |
|---|---|---|
| Predicted: fraud | 24TP | 6FP |
| Predicted: genuine transaction | 16FN | 54TN |
Accuracy = (TP + TN) / Total = (24 + 54) / 100 = 78 / 100 = 0.78 (78%)
Precision = TP / (TP + FP) = 24 / (24 + 6) = 24 / 30 = 0.80 (80%)
Recall = TP / (TP + FN) = 24 / (24 + 16) = 24 / 40 = 0.60 (60%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.80 x 0.60 / (0.80 + 0.60) = 0.96 / 1.40 = 0.69
Recall, because a fraud that is missed (False Negative) causes a loss of money.
The confusion matrix of a plagiarism checker, tested on 200 cases, is given below. Calculate:
Confusion matrix
| Reality: copied work | Reality: original work | |
|---|---|---|
| Predicted: copied work | 56TP | 56FP |
| Predicted: original work | 44FN | 44TN |
Accuracy = (TP + TN) / Total = (56 + 44) / 200 = 100 / 200 = 0.50 (50%)
Precision = TP / (TP + FP) = 56 / (56 + 56) = 56 / 112 = 0.50 (50%)
Recall = TP / (TP + FN) = 56 / (56 + 44) = 56 / 100 = 0.56 (56%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.50 x 0.56 / (0.50 + 0.56) = 0.56 / 1.06 = 0.53
A face-unlock system (positive = owner) was tested on 1000 cases. It predicted 'owner' for 240 cases; out of these, 120 were really 'owner'. It predicted 'stranger' for the other 760 cases; out of these, 680 were really 'stranger'.
TP = 120, FP = 240 - 120 = 120, TN = 680, FN = 760 - 680 = 80.
Confusion matrix
| Reality: owner | Reality: stranger | |
|---|---|---|
| Predicted: owner | 120TP | 120FP |
| Predicted: stranger | 80FN | 680TN |
Accuracy = (TP + TN) / Total = (120 + 680) / 1000 = 800 / 1000 = 0.80 (80%)
Precision = TP / (TP + FP) = 120 / (120 + 120) = 120 / 240 = 0.50 (50%)
Recall = TP / (TP + FN) = 120 / (120 + 80) = 120 / 200 = 0.60 (60%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.50 x 0.60 / (0.50 + 0.60) = 0.6 / 1.10 = 0.55
Precision, because a stranger accepted as the owner (False Positive) can open the phone.
The confusion matrix of a flood-warning model, tested on 200 cases, is given below. Calculate:
Confusion matrix
| Reality: flood | Reality: no flood | |
|---|---|---|
| Predicted: flood | 36TP | 44FP |
| Predicted: no flood | 4FN | 116TN |
Accuracy = (TP + TN) / Total = (36 + 116) / 200 = 152 / 200 = 0.76 (76%)
Precision = TP / (TP + FP) = 36 / (36 + 44) = 36 / 80 = 0.45 (45%)
Recall = TP / (TP + FN) = 36 / (36 + 4) = 36 / 40 = 0.90 (90%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.45 x 0.90 / (0.45 + 0.90) = 0.81 / 1.35 = 0.60
An AI that finds pneumonia in chest X-rays was tested on 200 cases. It predicted 'pneumonia' for 48 cases; out of these, 36 were really 'pneumonia'. It predicted 'normal' for the other 152 cases; out of these, 116 were really 'normal'.
TP = 36, FP = 48 - 36 = 12, TN = 116, FN = 152 - 116 = 36.
Confusion matrix
| Reality: pneumonia | Reality: normal | |
|---|---|---|
| Predicted: pneumonia | 36TP | 12FP |
| Predicted: normal | 36FN | 116TN |
Accuracy = (TP + TN) / Total = (36 + 116) / 200 = 152 / 200 = 0.76 (76%)
Precision = TP / (TP + FP) = 36 / (36 + 12) = 36 / 48 = 0.75 (75%)
Recall = TP / (TP + FN) = 36 / (36 + 36) = 36 / 72 = 0.50 (50%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.75 x 0.50 / (0.75 + 0.50) = 0.75 / 1.25 = 0.60
Recall, because a patient whose pneumonia is missed (False Negative) goes untreated.
The confusion matrix of a filter that blocks unsafe videos for children, tested on 100 cases, is given below. Calculate:
Confusion matrix
| Reality: unsafe video | Reality: safe video | |
|---|---|---|
| Predicted: unsafe video | 54TP | 18FP |
| Predicted: safe video | 6FN | 22TN |
Accuracy = (TP + TN) / Total = (54 + 22) / 100 = 76 / 100 = 0.76 (76%)
Precision = TP / (TP + FP) = 54 / (54 + 18) = 54 / 72 = 0.75 (75%)
Recall = TP / (TP + FN) = 54 / (54 + 6) = 54 / 60 = 0.90 (90%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.75 x 0.90 / (0.75 + 0.90) = 1.35 / 1.65 = 0.82
A camera that finds defective bottles in a factory was tested on 1000 cases. It predicted 'defective bottle' for 200 cases; out of these, 160 were really 'defective bottle'. It predicted 'good bottle' for the other 800 cases; out of these, 640 were really 'good bottle'.
TP = 160, FP = 200 - 160 = 40, TN = 640, FN = 800 - 640 = 160.
Confusion matrix
| Reality: defective bottle | Reality: good bottle | |
|---|---|---|
| Predicted: defective bottle | 160TP | 40FP |
| Predicted: good bottle | 160FN | 640TN |
Accuracy = (TP + TN) / Total = (160 + 640) / 1000 = 800 / 1000 = 0.80 (80%)
Precision = TP / (TP + FP) = 160 / (160 + 40) = 160 / 200 = 0.80 (80%)
Recall = TP / (TP + FN) = 160 / (160 + 160) = 160 / 320 = 0.50 (50%)
F1 score = 2 x Precision x Recall / (Precision + Recall) = 2 x 0.80 x 0.50 / (0.80 + 0.50) = 0.8 / 1.30 = 0.62
Recall, because a defective bottle that is passed (False Negative) reaches the customer.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (9 words): aman, likes, to, play, cricket, riya, chess, and, together
Document vector table (frequency of each word)
| aman | likes | to | play | cricket | riya | chess | and | together | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 1 | 1 | 1 | 0 | 1 | 1 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 1 | 0 | 1 | 0 | 1 | 1 |
Document frequency and inverse document frequency
| Word | aman | likes | to | play | cricket | riya | chess | and | together |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 2 | 3 | 1 | 2 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/2 | 3/2 | 3/2 | 3/3 | 3/1 | 3/2 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.176 | 0.176 | 0.176 | 0.000 | 0.477 | 0.176 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (12 words): we, love, our, school, has, a, big, library, read, books, in, the
Document vector table (frequency of each word)
| we | love | our | school | has | a | big | library | read | books | in | the | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 | 1 |
| Word | we | love | our | school | has | a | big | library | read | books | in | the |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 1 | 2 | 2 | 1 | 1 | 1 | 2 | 1 | 1 | 1 | 1 |
TF of 'love' in Document 1 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
The words that occur in only one document (DF = 1), such as love, has, a, big, read, books, in. They have the highest IDF (3/1), so their TF-IDF value is the highest.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (10 words): the, sun, is, hot, moon, cool, and, are, in, sky
Document vector table (frequency of each word)
| the | sun | is | hot | moon | cool | and | are | in | sky | |
|---|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 1 | 0 | 1 | 0 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 3 | 3 | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 1 | 1 |
Document frequency and inverse document frequency
| Word | the | sun | is | hot | moon | cool | and | are | in | sky |
|---|---|---|---|---|---|---|---|---|---|---|
| DF | 3 | 2 | 2 | 1 | 2 | 1 | 1 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/3 | 3/2 | 3/2 | 3/1 | 3/2 | 3/1 | 3/1 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.000 | 0.176 | 0.176 | 0.477 | 0.176 | 0.477 | 0.477 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (9 words): rain, helps, the, farmers, grow, rice, in, we, eat
Document vector table (frequency of each word)
| rain | helps | the | farmers | grow | rice | in | we | eat | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 1 | 0 | 0 | 1 | 1 | 1 | 1 | 0 | 0 |
| Document 3 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | 1 |
| Word | rain | helps | the | farmers | grow | rice | in | we | eat |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 1 | 1 | 2 | 1 | 2 | 1 | 1 | 1 |
TF of 'helps' in Document 1 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
The words that occur in only one document (DF = 1), such as helps, the, grow, in, we. They have the highest IDF (3/1), so their TF-IDF value is the highest.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (9 words): i, like, mango, juice, is, the, king, of, fruits
Document vector table (frequency of each word)
| i | like | mango | juice | is | the | king | of | fruits | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 1 | 0 | 1 | 1 | 1 | 1 | 1 |
| Document 3 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 |
Document frequency and inverse document frequency
| Word | i | like | mango | juice | is | the | king | of | fruits |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 2 | 1 | 1 | 1 | 1 | 1 | 2 |
| IDF = 3 / DF | 3/2 | 3/2 | 3/2 | 3/1 | 3/1 | 3/1 | 3/1 | 3/1 | 3/2 |
| log(IDF) | 0.176 | 0.176 | 0.176 | 0.477 | 0.477 | 0.477 | 0.477 | 0.477 | 0.176 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (8 words): data, is, the, new, oil, ai, needs, future
Document vector table (frequency of each word)
| data | is | the | new | oil | ai | needs | future | |
|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 |
| Document 2 | 1 | 0 | 0 | 0 | 0 | 1 | 1 | 0 |
| Document 3 | 0 | 1 | 1 | 0 | 0 | 1 | 0 | 1 |
| Word | data | is | the | new | oil | ai | needs | future |
|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 2 | 1 | 1 | 2 | 1 | 1 |
TF of 'new' in Document 1 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
The words that occur in only one document (DF = 1), such as new, oil, needs. They have the highest IDF (3/1), so their TF-IDF value is the highest.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (8 words): cats, drink, milk, dogs, water, and, are, pets
Document vector table (frequency of each word)
| cats | drink | milk | dogs | water | and | are | pets | |
|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 1 | 0 | 1 | 1 | 0 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 1 |
Document frequency and inverse document frequency
| Word | cats | drink | milk | dogs | water | and | are | pets |
|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 1 | 2 | 1 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/2 | 3/2 | 3/1 | 3/2 | 3/1 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.176 | 0.176 | 0.477 | 0.176 | 0.477 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (11 words): trees, give, us, oxygen, we, must, plant, and, save, the, earth
Document vector table (frequency of each word)
| trees | give | us | oxygen | we | must | plant | and | save | the | earth | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 1 | 0 | 0 | 0 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 | 1 |
| Word | trees | give | us | oxygen | we | must | plant | and | save | the | earth |
|---|---|---|---|---|---|---|---|---|---|---|---|
| DF | 3 | 1 | 1 | 1 | 1 | 1 | 2 | 1 | 1 | 1 | 1 |
TF of 'give' in Document 1 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
trees. It occurs in all 3 documents, so IDF = 3/3 = 1 and log(1) = 0. Such a word does not help to tell the documents apart.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (10 words): ravi, plays, football, daily, is, a, team, game, loves, his
Document vector table (frequency of each word)
| ravi | plays | football | daily | is | a | team | game | loves | his | |
|---|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 1 | 0 | 1 | 1 | 1 | 1 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | 1 |
Document frequency and inverse document frequency
| Word | ravi | plays | football | daily | is | a | team | game | loves | his |
|---|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 1 | 2 | 1 | 1 | 1 | 2 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/2 | 3/1 | 3/2 | 3/1 | 3/1 | 3/1 | 3/2 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.176 | 0.477 | 0.176 | 0.477 | 0.477 | 0.477 | 0.176 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (9 words): reading, is, a, good, habit, makes, life, books, fun
Document vector table (frequency of each word)
| reading | is | a | good | habit | makes | life | books | fun | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 2 | 2 | 1 | 1 | 1 | 0 | 0 |
| Document 3 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 1 | 1 |
| Word | reading | is | a | good | habit | makes | life | books | fun |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 2 | 2 | 2 | 1 | 1 | 1 | 1 |
TF of 'makes' in Document 2 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
The words that occur in only one document (DF = 1), such as makes, life, books. They have the highest IDF (3/1), so their TF-IDF value is the highest.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (10 words): divya, went, to, the, market, was, closed, came, back, home
Document vector table (frequency of each word)
| divya | went | to | the | market | was | closed | came | back | home | |
|---|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 0 | 1 | 1 | 1 | 1 | 0 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 |
Document frequency and inverse document frequency
| Word | divya | went | to | the | market | was | closed | came | back | home |
|---|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 1 | 1 | 2 | 2 | 1 | 1 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/2 | 3/1 | 3/1 | 3/2 | 3/2 | 3/1 | 3/1 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.176 | 0.477 | 0.477 | 0.176 | 0.176 | 0.477 | 0.477 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (8 words): robots, can, help, doctors, sick, people, not, feel
Document vector table (frequency of each word)
| robots | can | help | doctors | sick | people | not | feel | |
|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 1 | 1 | 1 | 1 | 0 | 0 |
| Document 3 | 1 | 1 | 0 | 0 | 0 | 0 | 1 | 1 |
| Word | robots | can | help | doctors | sick | people | not | feel |
|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 2 | 2 | 1 | 1 | 1 | 1 |
TF of 'sick' in Document 2 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
The words that occur in only one document (DF = 1), such as sick, people, not. They have the highest IDF (3/1), so their TF-IDF value is the highest.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (7 words): clean, water, is, life, save, every, day
Document vector table (frequency of each word)
| clean | water | is | life | save | every | day | |
|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 |
| Document 2 | 0 | 1 | 0 | 0 | 1 | 1 | 1 |
| Document 3 | 0 | 1 | 1 | 1 | 0 | 0 | 0 |
Document frequency and inverse document frequency
| Word | clean | water | is | life | save | every | day |
|---|---|---|---|---|---|---|---|
| DF | 1 | 3 | 2 | 2 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/1 | 3/3 | 3/2 | 3/2 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.477 | 0.000 | 0.176 | 0.176 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (9 words): yoga, keeps, the, body, fit, a, has, mind, calm
Document vector table (frequency of each word)
| yoga | keeps | the | body | fit | a | has | mind | calm | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 0 | 1 | 2 | 2 | 1 | 1 | 0 |
| Document 3 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 1 | 1 |
| Word | yoga | keeps | the | body | fit | a | has | mind | calm |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 2 | 2 | 2 | 1 | 1 | 2 | 1 |
TF of 'a' in Document 2 = 2. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 2 x log(3/1) = 2 x 0.477 = 0.954
The words that occur in only one document (DF = 1), such as a, has. They have the highest IDF (3/1), so their TF-IDF value is the highest.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (9 words): the, bus, was, late, today, i, missed, for, school
Document vector table (frequency of each word)
| the | bus | was | late | today | i | missed | for | school | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
| Document 2 | 1 | 1 | 0 | 0 | 0 | 1 | 1 | 0 | 0 |
| Document 3 | 0 | 0 | 1 | 1 | 1 | 1 | 0 | 1 | 1 |
Document frequency and inverse document frequency
| Word | the | bus | was | late | today | i | missed | for | school |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 2 | 2 | 2 | 2 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/2 | 3/2 | 3/2 | 3/2 | 3/2 | 3/2 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.176 | 0.176 | 0.176 | 0.176 | 0.176 | 0.176 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (9 words): honey, bees, make, live, in, a, hive, is, sweet
Document vector table (frequency of each word)
| honey | bees | make | live | in | a | hive | is | sweet | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 2 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 1 | 0 | 1 | 1 | 1 | 1 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 |
| Word | honey | bees | make | live | in | a | hive | is | sweet |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
TF of 'make' in Document 1 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
The words that occur in only one document (DF = 1), such as make, live, in, a, hive, is. They have the highest IDF (3/1), so their TF-IDF value is the highest.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (9 words): solar, panels, save, power, we, at, home, is, clean
Document vector table (frequency of each word)
| solar | panels | save | power | we | at | home | is | clean | |
|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 1 | 1 | 1 | 1 | 1 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 1 | 0 | 0 | 0 | 1 | 1 |
Document frequency and inverse document frequency
| Word | solar | panels | save | power | we | at | home | is | clean |
|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 1 | 2 | 3 | 1 | 1 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/2 | 3/1 | 3/2 | 3/3 | 3/1 | 3/1 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.176 | 0.477 | 0.176 | 0.000 | 0.477 | 0.477 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (10 words): chess, is, a, mind, game, asha, plays, well, has, sharp
Document vector table (frequency of each word)
| chess | is | a | mind | game | asha | plays | well | has | sharp | |
|---|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 1 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 0 | 0 |
| Document 3 | 0 | 0 | 1 | 1 | 0 | 1 | 0 | 0 | 1 | 1 |
| Word | chess | is | a | mind | game | asha | plays | well | has | sharp |
|---|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 1 | 2 | 2 | 1 | 2 | 1 | 1 | 1 | 1 |
TF of 'is' in Document 1 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
The words that occur in only one document (DF = 1), such as is, game, plays, well, has. They have the highest IDF (3/1), so their TF-IDF value is the highest.
Consider the corpus of three documents given below. After text normalisation (change to lower case and remove punctuation; keep all the words),:
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (12 words): my, phone, has, a, camera, the, takes, clear, photos, i, share, on
Document vector table (frequency of each word)
| my | phone | has | a | camera | the | takes | clear | photos | i | share | on | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 |
| Document 3 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 | 1 | 1 |
Document frequency and inverse document frequency
| Word | my | phone | has | a | camera | the | takes | clear | photos | i | share | on |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 2 | 1 | 1 | 2 | 1 | 1 | 1 | 2 | 1 | 1 | 1 |
| IDF = 3 / DF | 3/2 | 3/2 | 3/1 | 3/1 | 3/2 | 3/1 | 3/1 | 3/1 | 3/2 | 3/1 | 3/1 | 3/1 |
| log(IDF) | 0.176 | 0.176 | 0.477 | 0.477 | 0.176 | 0.477 | 0.477 | 0.477 | 0.176 | 0.477 | 0.477 | 0.477 |
TF-IDF(W) = TF(W) x log(IDF(W)), where TF is the frequency of the word in the document and IDF = total number of documents / document frequency. The log is to the base 10.
Consider the corpus of three documents given below. Apply the Bag of Words algorithm after text normalisation (change to lower case and remove punctuation; keep all the words).
Corpus
Rule used: all the letters are changed to lower case and the punctuation marks are removed; no word is dropped.
Vocabulary (10 words): india, won, the, final, match, was, in, mumbai, plays, well
Document vector table (frequency of each word)
| india | won | the | final | match | was | in | mumbai | plays | well | |
|---|---|---|---|---|---|---|---|---|---|---|
| Document 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 |
| Document 2 | 0 | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 0 | 0 |
| Document 3 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 1 |
| Word | india | won | the | final | match | was | in | mumbai | plays | well |
|---|---|---|---|---|---|---|---|---|---|---|
| DF | 2 | 1 | 2 | 2 | 2 | 1 | 1 | 1 | 1 | 1 |
TF of 'won' in Document 1 = 1. DF = 1, so IDF = 3/1. TF-IDF = TF x log(IDF) = 1 x log(3/1) = 1 x 0.477 = 0.477
The words that occur in only one document (DF = 1), such as won, was, in, mumbai, plays. They have the highest IDF (3/1), so their TF-IDF value is the highest.
A single neuron (perceptron) is used to decide: "Should Riya go to the park?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| The weather is good | 3 | Yes (1) | No (0) |
| Her friend is coming | 5 | No (0) | Yes (1) |
| Her homework is done | 1 | No (0) | Yes (1) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (1 x 3) + (0 x 5) + (0 x 1) + (-1) = 2. It is less than 4, so the output is 0 (No).
Sum = (0 x 3) + (1 x 5) + (1 x 1) + (-1) = 5. It is not less than 4, so the output is 1 (Yes).
Her friend is coming, because it has the largest weight (5). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should Arjun buy the phone?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -1. The output is 1 (Yes) if the weighted sum is 3 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| The price is within budget | 2 | No (0) | Yes (1) |
| The camera is good | 1 | Yes (1) | No (0) |
| The reviews are good | 5 | Yes (1) | No (0) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (0 x 2) + (1 x 1) + (1 x 5) + (-1) = 5. It is not less than 3, so the output is 1 (Yes).
Sum = (1 x 2) + (0 x 1) + (0 x 5) + (-1) = 1. It is less than 3, so the output is 0 (No).
The reviews are good, because it has the largest weight (5). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should the school team play the match?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| The ground is dry | 3 | Yes (1) | Yes (1) |
| All the players are fit | 5 | Yes (1) | No (0) |
| The coach is present | 1 | Yes (1) | No (0) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (1 x 3) + (1 x 5) + (1 x 1) + (-1) = 8. It is not less than 4, so the output is 1 (Yes).
Sum = (1 x 3) + (0 x 5) + (0 x 1) + (-1) = 2. It is less than 4, so the output is 0 (No).
All the players are fit, because it has the largest weight (5). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should Meena watch the film?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is 1. The output is 1 (Yes) if the weighted sum is 6 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| Her exam is over | 1 | Yes (1) | No (0) |
| Her friends are going | 5 | No (0) | Yes (1) |
| The ticket is cheap | 4 | Yes (1) | No (0) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (1 x 1) + (0 x 5) + (1 x 4) + (1) = 6. It is not less than 6, so the output is 1 (Yes).
Sum = (0 x 1) + (1 x 5) + (0 x 4) + (1) = 6. It is not less than 6, so the output is 1 (Yes).
Her friends are going, because it has the largest weight (5). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should the farmer water the field today?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -2. The output is 1 (Yes) if the weighted sum is 3 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| The soil is dry | 1 | Yes (1) | No (0) |
| No rain is forecast | 5 | No (0) | Yes (1) |
| The pump has power | 3 | No (0) | Yes (1) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (1 x 1) + (0 x 5) + (0 x 3) + (-2) = -1. It is less than 3, so the output is 0 (No).
Sum = (0 x 1) + (1 x 5) + (1 x 3) + (-2) = 6. It is not less than 3, so the output is 1 (Yes).
No rain is forecast, because it has the largest weight (5). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should Kabir join the coding club?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is 1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| He is free on that day | 3 | Yes (1) | No (0) |
| He likes coding | 1 | No (0) | Yes (1) |
| His friend has joined | 2 | No (0) | Yes (1) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (1 x 3) + (0 x 1) + (0 x 2) + (1) = 4. It is not less than 4, so the output is 1 (Yes).
Sum = (0 x 3) + (1 x 1) + (1 x 2) + (1) = 4. It is not less than 4, so the output is 1 (Yes).
He is free on that day, because it has the largest weight (3). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should the family go on a picnic?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| It is a holiday | 2 | No (0) | Yes (1) |
| The weather is clear | 4 | Yes (1) | No (0) |
| The car is available | 1 | Yes (1) | No (0) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (0 x 2) + (1 x 4) + (1 x 1) + (-1) = 4. It is not less than 4, so the output is 1 (Yes).
Sum = (1 x 2) + (0 x 4) + (0 x 1) + (-1) = 1. It is less than 4, so the output is 0 (No).
The weather is clear, because it has the largest weight (4). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should Sana order food online?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is 1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| She is very hungry | 4 | Yes (1) | No (0) |
| There is a discount | 1 | Yes (1) | No (0) |
| Delivery is fast | 2 | No (0) | Yes (1) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (1 x 4) + (1 x 1) + (0 x 2) + (1) = 6. It is not less than 4, so the output is 1 (Yes).
Sum = (0 x 4) + (0 x 1) + (1 x 2) + (1) = 3. It is less than 4, so the output is 0 (No).
She is very hungry, because it has the largest weight (4). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should the shop open on Sunday?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| Many customers are expected | 2 | Yes (1) | No (0) |
| Staff is available | 3 | Yes (1) | No (0) |
| It is a festival week | 4 | No (0) | Yes (1) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (1 x 2) + (1 x 3) + (0 x 4) + (-1) = 4. It is not less than 4, so the output is 1 (Yes).
Sum = (0 x 2) + (0 x 3) + (1 x 4) + (-1) = 3. It is less than 4, so the output is 0 (No).
It is a festival week, because it has the largest weight (4). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should Dev go for the morning run?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| He slept well | 1 | No (0) | Yes (1) |
| It is not raining | 4 | No (0) | Yes (1) |
| His shoes are dry | 5 | Yes (1) | No (0) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (0 x 1) + (0 x 4) + (1 x 5) + (-1) = 4. It is not less than 4, so the output is 1 (Yes).
Sum = (1 x 1) + (1 x 4) + (0 x 5) + (-1) = 4. It is not less than 4, so the output is 1 (Yes).
His shoes are dry, because it has the largest weight (5). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should the class go on the science trip?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| Parents have agreed | 4 | No (0) | Yes (1) |
| The bus is booked | 5 | Yes (1) | No (0) |
| The museum is open | 1 | Yes (1) | No (0) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (0 x 4) + (1 x 5) + (1 x 1) + (-1) = 5. It is not less than 4, so the output is 1 (Yes).
Sum = (1 x 4) + (0 x 5) + (0 x 1) + (-1) = 3. It is less than 4, so the output is 0 (No).
The bus is booked, because it has the largest weight (5). A larger weight means the input has more effect on the decision.
A single neuron (perceptron) is used to decide: "Should Tara adopt the puppy?" Each factor is an input: Yes = 1, No = 0. The weights are given in the table. The bias is -1. The output is 1 (Yes) if the weighted sum is 4 or more, otherwise 0 (No).
| Factor (input) | Weight | Day 1 | Day 2 |
|---|---|---|---|
| She has time for it | 2 | Yes (1) | No (0) |
| Her parents agree | 5 | No (0) | Yes (1) |
| There is space at home | 1 | Yes (1) | No (0) |
Weighted sum = (x1 x w1) + (x2 x w2) + (x3 x w3) + bias
Sum = (1 x 2) + (0 x 5) + (1 x 1) + (-1) = 2. It is less than 4, so the output is 0 (No).
Sum = (0 x 2) + (1 x 5) + (0 x 1) + (-1) = 4. It is not less than 4, so the output is 1 (Yes).
Her parents agree, because it has the largest weight (5). A larger weight means the input has more effect on the decision.
Explain the communication cycle with a workplace example.
Sender - a sales manager has an idea (weekly target); Encoding - writes 'Target: 50 units this week'; Channel - e-mail; Receiver - the sales team; Decoding - team understands the target; Feedback - a member asks whether online orders count. Feedback completes the cycle and confirms understanding.
Differentiate between oral and written communication (any four points).
Oral uses spoken words (meetings, calls); written uses text (letters, e-mails). Oral is fast with immediate feedback; written feedback is slow. Oral has no permanent record; written gives a record and proof. Oral uses tone to add meaning; written needs good language skills and can be planned and edited.
What are the qualities of good feedback? Give an example of specific feedback.
Good feedback is specific, timely, polite, focused on the work not the person, balanced (good points and improvements) and actionable. Example: 'Your data is correct; add labels to the graph and write a conclusion.' It tells exactly what to keep and what to change.
Explain any four barriers to effective communication with solutions.
Physical - noise, poor network; use a quiet place and good equipment. Linguistic - different language, jargon; use simple words or a translator. Interpersonal - lack of trust, anger; stay calm and listen. Organisational - many levels, strict rules; clear channels and open-door policy. (Cultural and psychological barriers may also be written.)
Change into passive voice:
(i) Trees were planted by students. (ii) The letter was signed by the manager. (iii) Poems are written by Riya. (iv) The shop will be opened by them.
What is stress? Write its signs and any four ways to manage it.
Stress is the feeling of pressure when facing a difficult situation. Signs: headache, poor sleep, irritation, avoiding work. Ways: time management, physical exercise, yoga and deep breathing, healthy food and sleep, talking to family and friends, hobbies and positive thinking. (Any four.)
Explain the four steps of time management.
Organise - list all tasks using a diary or app. Prioritise - do the most important and urgent work first. Control - avoid distractions like social media and say no politely. Track - check at the end of the day what was done and what is left.
Make a SMART goal for improving fitness and explain each letter.
'I will jog 20 minutes every morning, 5 days a week, and reduce my 1 km time from 8 to 6 minutes by 31 March.' Specific - jogging; Measurable - 20 minutes and running time; Achievable - possible daily; Realistic - fits my routine; Time-bound - by 31 March.
Explain the urgent-important matrix with examples.
Important and urgent - do now (assignment due tomorrow). Important but not urgent - plan (revision for next month's exam). Urgent but not important - delegate or do quickly (some calls). Neither urgent nor important - avoid (endless scrolling). It helps us prioritise.
Explain the ABC model of stress with an example.
A - Adversity, the event (low marks in a test). B - Belief, our thought about it ('I am a failure'). C - Consequence, our feeling and action (sadness, stops studying). Changing the belief to 'I need more practice' changes the consequence to extra practice and better marks.
What is an operating system? Explain any four functions.
An operating system is system software that manages hardware and software and acts as an interface between the user and the computer (Windows, Linux, Android). Functions: process management, memory management, file management, device management through drivers, security with user accounts and passwords, and providing a user interface. (Any four.)
Differentiate between GUI and CLI.
GUI (Graphical User Interface) uses icons, windows and menus operated with a mouse or touch; it is easy for beginners (Windows, Android). CLI (Command Line Interface) needs typed text commands using the keyboard; it is faster for experts but harder to learn (Linux terminal, MS-DOS).
Explain virus, worm, Trojan horse and ransomware.
Virus - attaches to files and harms data when the file runs. Worm - copies itself and spreads automatically through networks. Trojan horse - looks useful but harms when run; does not copy itself. Ransomware - locks or encrypts files and demands money to unlock them.
What is phishing? How can we protect ourselves from it?
Phishing is a fake e-mail, SMS or website pretending to be a trusted organisation to steal passwords, OTPs or bank details. Protection: never share OTP/PIN; check the sender and URL; do not click unknown links; use two-factor authentication; report it to cybercrime.gov.in or helpline 1930.
Write any four computer care tasks and their purpose.
Clean the keyboard and vents - prevents dust damage and overheating. Install updates - fixes security holes. Run antivirus scans - removes malware. Back up data - protects against loss. Delete unwanted files - keeps the computer fast. Use a UPS - protects from power cuts.
Explain the role of entrepreneurship in society.
Entrepreneurship creates jobs and self-employment, brings new products and services, lowers prices through competition, develops villages and small towns, increases exports and national income, and solves social problems through social enterprises, e.g. low-cost sanitary pad machines run by rural women.
Explain the five functions of an entrepreneur with an example.
Decision making - what, where and how to produce; managing - planning and controlling daily work; dividing income - paying wages, rent and suppliers; taking risk - investing without certainty of profit; innovation - new methods. Example: a bakery owner decides on eggless cakes, hires bakers, pays staff, takes a loan for an oven and starts online orders.
Differentiate between wage employment and self-employment.
Wage employment means working for others for a fixed salary with less risk, following the employer's decisions. Self-employment means running one's own business; income depends on profit, there is more risk, but there is freedom to decide and the chance to create jobs for others.
Explain the stages of the entrepreneurship process.
Self-analysis - know interests and capacity for risk; identify an opportunity - unmet needs; generate and select an idea; prepare a business plan - product, customers, costs, profit; arrange finance - savings, loans, schemes; launch small and test; manage and grow.
A tailor buys cloth for Rs 8,000 and spends Rs 1,500 on thread and electricity. She sells shirts for Rs 13,000. Find the total cost and profit, and name two government schemes that help such entrepreneurs.
Total cost = 8,000 + 1,500 = Rs 9,500. Profit = 13,000 - 9,500 = Rs 3,500. Schemes: PM Mudra Yojana, Startup India, Stand-Up India, Skill India. (Any two.)
What is sustainable development? Explain its three dimensions.
Sustainable development meets the needs of the present without compromising the ability of future generations to meet their own needs. Economic - jobs, income and growth; Social - education, health, equality; Environmental - clean air, water and biodiversity. All three must be balanced.
Explain any four problems related to sustainable development and their solutions.
Water shortage - rainwater harvesting, drip irrigation. Energy crisis - solar, wind, biogas. Soil degradation - organic farming, crop rotation. Air pollution - public transport, EVs. Waste - reduce, reuse, recycle and segregate.
Name any eight Sustainable Development Goals.
No poverty; zero hunger; good health and well-being; quality education; gender equality; clean water and sanitation; affordable and clean energy; decent work and economic growth; climate action; life below water; life on land. (Any eight.)
What are green jobs? Give four examples.
Green jobs are jobs that help protect or restore the environment. Examples: solar technician, EV mechanic, recycling worker, organic farmer, green building designer, water conservationist. (Any four.)
Explain how students can contribute to sustainable development at home, at school and in society.
At home - turn off taps, lights and fans; use cloth bags; repair before replacing. At school - eco-club, composting pit, both sides of paper, no food waste. In society - cleanliness and plantation drives, public transport, support local products, spread awareness.
Explain the difference between grayscale and RGB images. How many values are needed to store a 100 x 50 colour image?
A grayscale image has one value per pixel from 0 (black) to 255 (white). An RGB image has three values per pixel - red, green and blue - each 0 to 255, mixed to make colours. A 100 x 50 image has 5,000 pixels; in RGB it needs 3 x 5,000 = 15,000 values.
Explain the four computer-vision tasks with examples.
Image classification - one label for the whole image (dog). Classification + localisation - label and a box for a single object. Object detection - labels and boxes for many objects (cars and people in a street). Instance segmentation - exact pixel outline of each object (each person in a crowd).
Explain the layers of a convolutional neural network.
Convolution layer - applies kernels to find features and makes feature maps; ReLU layer - replaces negative values with zero; Pooling layer - reduces the size of feature maps (e.g. max pooling) and makes the network faster; Fully connected layer - combines features and gives final class probabilities.
Perform convolution of the 3 x 3 image patch [[1, 2, 1], [0, 0, 0], [3, 3, 3]] with the kernel [[1, 1, 1], [0, 0, 0], [-1, -1, -1]]. What does the result indicate?
Multiply element by element and add: (1 + 2 + 1) + 0 + (-3 - 3 - 3) = 4 - 9 = -5. A large magnitude indicates a horizontal edge, i.e. a change in brightness from the top row to the bottom row.
Write any four applications of computer vision and two ethical concerns.
Face unlock and attendance; self-driving cars; medical image analysis; number-plate recognition at tolls; crop disease detection; Google Lens. Concerns: privacy and surveillance without consent; bias against some groups; deepfakes.
Tap an option to check your answer at once; your score is saved
Ten multiple-choice questions picked at random. Tap an option to answer. You get your score at the end.
Ten ready papers with blueprints and a random paper maker, all on the design of the CBSE Sample Question Paper 2026-27
Each paper has 21 questions for 50 marks and 2 hours, with the same choices as the CBSE Sample Question Paper 2026-27: any 4 of 6 in Q1, any 5 of 6 in Q2 to Q5, any 3 of Q6 to Q10, any 4 of Q11 to Q16 and any 3 of Q17 to Q21. Every paper has numericals in Section A, and a confusion matrix, a Bag of Words and a neuron numerical in Q17 to Q21. No question is repeated: the ten papers use 460 different questions.
Downloads and printing work when this page is opened as an HTML file in a browser.
Choose Paper 1 to Paper 10, or make a random paper. The blueprint and the paper open here.
Read this the evening before and again at the exam centre gate