November 10, 2025 by Rezyo Staff Writer AI, CBSE, CBSE Class 10 0 comments

CBSE Class 10 AI MCQs | Interactive AI Quiz, Notes & Scenarios

Welcome to the ultimate CBSE AI Multiple-Choice Hub by Rezyo.in. Prepare for your 2025 exams by testing your knowledge with our comprehensive, interactive quiz. This hub covers every unit of the CBSE AI curriculum, from core concepts in Data Science, Computer Vision, and Natural Language Processing (NLP) to advanced topics like neural networks, evaluation metrics (F1 Score, Precision, Recall), and AI ethics. Challenge yourself with real-world interview scenarios and practical Python code snippets to build a deep, practical understanding of Artificial Intelligence.

CBSE AI MCQ Hub – Rezyo.in

CBSE AI Multiple-Choice Hub

Test your knowledge with interactive MCQs, scenarios, and code.

Updated: October 2025

Easy Track: Core Concepts

1. Which AI domain deals with helping machines understand human language?

Explanation: Natural Language Processing (NLP) is the specific domain focused on the interaction between computers and human language, including tasks like translation, sentiment analysis, and chatbots.

2. What is the correct order of the first three steps in the AI Project Cycle?

Explanation: The AI Project Cycle begins with clearly defining the problem (Problem Scoping), then collecting the necessary data (Data Acquisition), and then understanding that data (Data Exploration).

3. A model incorrectly predicts a fire has started (a “false alarm”). In a confusion matrix, what is this called?

Explanation: A False Positive (FP) occurs when the model predicts a positive outcome (“Yes, there is a fire”) but the reality is negative (“No, there is no fire”). This is a false alarm.

4. What is AI Bias?

Explanation: AI Bias happens when a model’s decisions are prejudiced. This is often caused by training data that contains human biases (e.g., if a hiring model was trained on data where only men were hired, it might learn to unfairly reject women).

5. You give an AI a folder of 1000 unlabeled images and ask it to find *groups* of similar images. What type of learning is this?

Explanation: This is Unsupervised Learning. Because the data is “unlabeled” (you didn’t tell the AI which images are cats, dogs, etc.), the AI must find the patterns on its own. Grouping items is a task called “clustering”.

6. What is the main purpose of the “Data Exploration” step in the AI Project Cycle?

Explanation: Data Exploration is all about visualizing the data (using graphs, charts) to understand it. This helps you find patterns, identify errors (like outliers or missing values), and decide which type of model to use next.

7. A “smart assistant” like Siri or Alexa is a primary example of which AI domain?

Explanation: Smart assistants primarily work by understanding and responding to human speech. This process of understanding language is the core task of Natural Language Processing.

8. In AI, what is a “model”?

Explanation: A model is the ‘brain’ of the AI. It’s an algorithm that has learned patterns from a large dataset. Once trained, it can make predictions or decisions on new, unseen data.

9. What is Supervised Learning?

Explanation: “Supervised” means the training data is “labeled.” You show the model an image of a cat *and* the label “cat”. It learns by comparing its prediction to the correct answer you provide.

10. What is the goal of the “Data Acquisition” step?

Explanation: Data Acquisition is the second step in the AI Project Cycle. It is the process of collecting or gathering the data (from surveys, databases, sensors, etc.) that will be used for training and testing.

11. In Computer Vision, what is a “pixel”?

Explanation: A digital image is a grid of tiny dots called pixels. A computer stores the image as a set of numbers, where each number (or set of numbers) describes the color and brightness of one pixel.

12. What is the purpose of “Tokenization” in NLP?

Explanation: Tokenization is the first step in text normalization. It takes a full sentence (e.g., “The cat sat.”) and breaks it into smaller pieces or “tokens” (e.g., [“The”, “cat”, “sat”, “.”]).

Difficult Track: In-Depth Topics

1. What is the primary purpose of the “Pooling Layer” in a Convolutional Neural Network (CNN)?

Explanation: The Pooling Layer (like Max Pooling) downsamples the feature map, making it smaller. This reduces the number of parameters, which speeds up processing and helps make the model less sensitive to the exact location of features.

2. A word appears frequently in one document but is very rare in all other documents. What will its TF-IDF score be?

Explanation: The score will be high. The “Term Frequency (TF)” will be high because it appears often in the document. The “Inverse Document Frequency (IDF)” will also be high because it is rare in other documents. High TF * High IDF = High TF-IDF score.

3. Why is F1 Score used for “unbalanced” datasets (e.g., 99% “No” samples, 1% “Yes” samples)?

Explanation: In an unbalanced dataset, a model that just predicts the majority class (e.g., “No”) would have high Accuracy (99%) but be useless. F1 Score balances Precision and Recall, forcing the model to be good at finding the rare “Yes” cases (Recall) without raising too many false alarms (Precision).

4. In the K-Nearest Neighbour (KNN) algorithm, what generally happens if you set “K” to a very low value, like K=1?

Explanation: A K of 1 means the model *only* looks at the single closest neighbor. If that one neighbor is an outlier or an error, the prediction will be wrong. A higher K value (like 5 or 7) is more stable because it takes a “vote” from multiple neighbors.

5. In Computer Vision, what is a “kernel” in a Convolution Layer used for?

Explanation: A kernel is a small matrix of numbers (e.g., 3×3) that slides over the main image. It performs a calculation at each position to create a “feature map,” which highlights specific features like horizontal edges, vertical edges, or corners.

6. A model is designed to find *all* positive cases (like a medical test). It’s okay if it has some false alarms, but it *cannot* miss any real cases. Which metric should be maximized?

Explanation: Recall (calculated as TP / (TP + FN)) measures the model’s ability to find *all* actual positive cases. Maximizing recall means minimizing False Negatives (FN), which are the “missed” cases. This is critical in medical screening.

7. What is a major *limitation* of the “Bag of Words” (BoW) model in NLP?

Explanation: Bag of Words only counts the *frequency* of words (how many times each word appears). The sentences “man bites dog” and “dog bites man” have the exact same “bag of words” but completely different meanings. BoW ignores grammar and word order.

8. What is “overfitting” in model training?

Explanation: Overfitting occurs when a model learns the training data *too* well, including its noise and irrelevant details. It becomes like a student who memorizes the textbook but can’t answer a new question. This model will fail in the real world.

9. What is the main purpose of the ReLU (Rectified Linear Unit) layer in a CNN?

Explanation: ReLU is an “activation function”. Its job is to introduce non-linearity. Without it, the entire neural network would just be one big linear equation, which isn’t powerful enough to learn complex things like the shape of a face. ReLU does this by simply changing all negative pixel values to zero.

10. What is the core difference between a Classification and a Regression model?

Explanation: This is a fundamental concept. Classification *classifies* data into distinct groups (e.g., ‘Spam’/’Not Spam’, ‘Positive’/’Negative’). Regression *regresses* to a continuous value (e.g., predicting the price of a house, the temperature tomorrow).

11. What is the difference between Stemming and Lemmatization in NLP?

Explanation: Stemming is a crude, fast process (e.g., “studies” -> “studi”). Lemmatization is a slower, more intelligent process that uses a dictionary (e.g., “studies” -> “study”). Lemmatization gives a valid word, while stemming may not.

12. What is “underfitting” in model training?

Explanation: Underfitting is the opposite of overfitting. It happens when the model is not complex enough (e.g., trying to fit a curved line with a straight-line model). The model fails to learn the data and will perform poorly on *both* training and test data.

Interview Scenarios

Scenario 1: A client says their spam filter model has 98% accuracy. They want to know if this is good. What is the *best* follow-up question to ask?

Explanation: Spam filtering is a classic “unbalanced” problem (most emails are NOT spam). High accuracy is meaningless. You need to know the Precision (of all emails marked as spam, how many were *actually* spam?) and Recall (of all *actual* spam emails, how many did the filter catch?). F1 Score combines these two metrics.

Scenario 2: Your team is building a model to detect a rare, life-threatening disease. Which error is *worse* to make?

Explanation: A False Negative is far worse. Telling a sick person they are healthy means they won’t get treatment and the disease could be fatal. A False Positive is stressful (it’s a false alarm) and requires a follow-up test, but it is not life-threatening. In this case, you must optimize for high **Recall** (finding all the sick people).

Scenario 3: A company wants to predict the *exact price* (e.g., $350,150) a house will sell for. What *type* of AI model should they build?

Explanation: Regression models are used to predict a continuous numerical value (like a price, temperature, or height). A Classification model, in contrast, predicts a category (e.g., “Spam” or “Not Spam”, “Cat” or “Dog”).

Scenario 4: A team starts building an AI model before talking to the users. They spend 6 months on it. The final model works, but users don’t find it helpful. Which step of the AI Project Cycle did they fail at?

Explanation: Problem Scoping (using the 4Ws: Who, What, Where, Why) involves understanding *who* the stakeholders are and *why* they need the solution. By not talking to users, the team built a solution for the wrong problem, even if the technology worked.

Scenario 5: Your manager gives you a messy, 10GB dataset and asks you to “build a model.” What is the *first* thing you should do?

Explanation: You must *always* explore your data first. You need to understand what’s in the dataset, how much of it is missing or corrupt, and what patterns exist. Building a model without this step is like building a house with no blueprint.

Scenario 6: A user complains their new face-unlock AI doesn’t work in the dark. It worked perfectly in the well-lit office where it was trained. What is this problem called?

Explanation: This is a classic example of a biased dataset. The “Data Acquisition” step failed to collect data representing *all* real-world conditions (i.e., dark rooms). The model is “biased” towards well-lit images and fails on any data that looks different.

Scenario 7: A bank wants to group its customers into segments like “High Spenders” and “Bargain Hunters” to send different marketing emails, but they don’t have these labels. What type of model should they use?

Explanation: Since the data is “unlabeled” (they don’t know the groups yet), this is an Unsupervised Learning problem. The specific task of *finding* groups in data is called Clustering (e.g., K-Means Clustering).

Scenario 8: A model that predicts stock prices is 99% accurate when predicting *yesterday’s* prices, but fails when predicting *tomorrow’s* prices. What is this a classic example of?

Explanation: This is a perfect example of overfitting. The model has “memorized” the training data (yesterday’s prices) perfectly, but it failed to learn the *general pattern* of the market. As a result, it is useless on new, unseen data (tomorrow’s prices).

Code Snippets

Snippet 1: What is the goal of this NLP code?

words = ["running", "studies", "goes"]
lemmatized_words = [lemmatizer.lemmatize(w) for w in words]
# Output: ['run', 'study', 'go']

Explanation: The code is converting words to their dictionary root (lemma). “running” becomes “run”, “studies” becomes “study”, and “goes” becomes “go”. This is Lemmatization. Stemming would have just chopped the ends, resulting in “run”, “studi”, and “go”.

Snippet 2: A model is trained with this data. What problem will it likely have?

# Training Data
# 95% of images are of "dogs"
# 5% of images are of "cats"

Explanation: This is an unbalanced dataset. The model will learn a lot about “dogs” but very little about “cats”. This will create an AI Bias, where the model is very good at one class but fails on the other. It will likely predict “dog” for most cat images.

Snippet 3: What is this NLP code snippet doing?

text = "this is a very good example"
stopwords = ["is", "a", "very"]
clean_text = [word for word in text.split() if word not in stopwords]
# Output: ['this', 'good', 'example']

Explanation: The code splits a sentence into words (`text.split()`) and then keeps only the words that are *not* in the `stopwords` list. This process of removing common but low-value words is called Stopword Removal.

Snippet 4: What will be the output of this Python (NumPy) code?

import numpy as np

arr = np.array([10, 20, 30, 40, 50])
print(arr.mean())

Explanation: `np.array()` creates a NumPy array. The `.mean()` function calculates the average of the elements. (10 + 20 + 30 + 40 + 50) / 5 = 150 / 5 = 30.0.

Snippet 5: What is this Python (OpenCV) code snippet most likely doing?

import cv2

img = cv2.imread('image.jpg')
gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
cv2.imshow('Image', gray_img)

Explanation: `cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)` is the specific OpenCV function used to change an image’s colorspace from BGR (Blue, Green, Red) to GRAY (grayscale). `cv2.imshow` then displays the resulting gray image.

Snippet 6: What is this Python (Pandas) code used for?

import pandas as pd
df = pd.read_csv('data.csv')
print(df.head())

Explanation: The `.head()` function in Pandas is a key part of Data Exploration. It prints the first 5 rows of your DataFrame (the table) so you can quickly see what the columns are named and what the data looks like without loading the entire file.

Snippet 7: What is this Python (Pandas) code doing?

import pandas as pd
df = pd.read_csv('data.csv')
clean_df = df.dropna()
print(df.shape, clean_df.shape)

Explanation: `.dropna()` is a key data cleaning function in Pandas. It “drops” (deletes) any row from the table that contains a “NA” or “NaN” (Not a Number / missing) value. The code prints the shape (row/column count) before and after to show how many rows were removed.

Snippet 8: This model was evaluated. What do these metrics tell you?

# Model Evaluation
Accuracy: 99%
Precision: 100%
Recall: 5%
F1-Score: 9%

Explanation: This is a classic unbalanced dataset problem. The 5% Recall is the most important number. It means the model is *failing to find 95% of the real cases* (high False Negatives). The high Accuracy is misleading, and the low F1-Score confirms the model is not good.

Notes: Curriculum Summary

Unit 1: AI Basics & Ethics

AI aims to create smart machines. It is divided into three main domains. Understanding these domains helps us categorize AI projects.

Data Sciences

Analyzes numbers and text to find patterns and make forecasts.

Computer Vision

Teaches computers to see and interpret images and videos.

Natural Language Processing

Enables computers to work with human language (text or speech).

AI Ethics

Building AI responsibly is important. We must consider:

AI Bias: The model should be fair and not make prejudiced decisions based on its training data.
Data Privacy: The data used to train models must be protected and used ethically.
AI Access: Technology should be accessible to everyone and not create a wider digital divide.

Unit 2: AI Project Cycle

The AI Project Cycle is a 5-step framework for building any AI project from start to finish. It provides a clear path and structure for development.

Problem Scoping: Using the 4Ws (Who, What, Where, Why) to understand the problem completely.
Data Acquisition: Finding and collecting reliable and relevant data (e.g., from surveys, sensors, or websites).
Data Exploration: Visualizing the data to find patterns, trends, and errors before modelling.
Modelling: Choosing the right AI model (e.g., Classification, Regression) and training it with the data.
Evaluation: Testing the model with new data to check its performance using metrics like Accuracy or F1 Score.

Unit 3: Key Concepts Comparison

Understanding the differences between key terms is useful. Here are two common comparisons.

AI vs. Machine Learning vs. Deep Learning

Feature	Artificial Intelligence (AI)	Machine Learning (ML)	Deep Learning (DL)
Concept	Broad idea of smart machines	A part of AI; learns from data	A part of ML; uses neural networks
Example	A smart chatbot or robot	A spam filter in your email	A self-driving car’s vision system

Stemming vs. Lemmatization

Feature	Stemming	Lemmatization
Method	Chops off word endings	Uses dictionary rules
Example (from “studies”)	“studi” (not a real word)	“study” (a real word)
Speed	Fast	Slow

CBSE Class 10 AI MCQs | Interactive AI Quiz, Notes & Scenarios

CBSE Class 10 AI MCQs | Interactive AI Quiz, Notes & Scenarios

CBSE AI Multiple-Choice Hub

Easy Track: Core Concepts

Difficult Track: In-Depth Topics

Interview Scenarios

Code Snippets

Notes: Curriculum Summary

Unit 1: AI Basics & Ethics

Data Sciences

Computer Vision

Natural Language Processing

AI Ethics

Unit 2: AI Project Cycle

Unit 3: Key Concepts Comparison

AI vs. Machine Learning vs. Deep Learning

Stemming vs. Lemmatization

Charts & Visuals

Infographic: Evaluation Matrix

True Positive (TP)

False Positive (FP)

False Negative (FN)

True Negative (TN)

Interactive Charts

AI Project Cycle

Sample Model Evaluation

Related Posts

CBSE Class 10 Cyber Ethics Notes (Code 165) | Netiquettes, IPR & E-Commerce Security

AI Ethics for CBSE Class 10: Self-Driving Cars, Data Privacy, AI Bias, & Access

NLP (TF-IDF, BoW) & AI Ethics: CBSE Class 10 Chptr 10 AI Guide

CBSE Coordinate Geometry Class 10: Formulas Examples Demos