Chapter 3: Evaluating Models

Important questions — the ones the board asked most (14)

1. Q1Very short answer5 marksAsked: 2024

   4. Answer any 5 out of the given 6 questions : (i) Statement 1 : Confusion matrix is an evaluation metric. Statement 2 : Confusion Matrix is a record which helps in evaluation. (a) Both Statement 1 and Statement 2 are correct. (b) Both Statement 1 and Statement 2 are incorrect. (c) Statement 1 is correct and Statement 2 is incorrect. (d) Statement 2 is correct and Statement 1 is incorrect. (ii) Which form of unsupervised learning does the following diagram indicate ? (a) Clustering (b) Regression (c) Reinforcement learning (d) Classification (iii) Bag of Words is a ________ model which helps in extracting features out of the text which can be helpful in machine learning algorithms. (a) Data Science (DS) (b) Virtual Reality (VR) (c) Natural Language Processing (NLP) (d) Computer Vision (CV) (iv) Which of the following represents an example of a recommendation system ? (a) An online clothing store that offers a wide variety of clothing options. (b) A search engine that retrieves relevant web pages based on user queries. (c) An e-commerce website that displays customer reviews and ratings for products. (d) A music streaming platform that suggests songs and playlists based on user listening history. (v) Name any two search engines. (vi) What is the primary need for evaluating an AI model's performance in the AI Model Development process ? (a) To increase the complexity of the model. (b) To visualize the data. (c) To assess how well the chosen model will work in future. (d) To reduce the amount of data used for training.

   

   Medha Ma’am

   Ask Medha’s AI teacher

   Hear it from Medha Ma’am · 115 s

   Model answer · AI model evaluation, recommendation, NLP basics

   1. 1.(d) Statement 2 is correct and Statement 1 is incorrect. A confusion matrix is not an evaluation metric itself; it is a record of predictions versus reality that helps in evaluation.1 mark
   2. 2.(a) Clustering. The unlabelled data points are grouped into two clusters according to their similarity.1 mark
   3. 3.(c) Natural Language Processing (NLP).1 mark
   4. 4.(d) A music streaming platform that suggests songs and playlists based on user listening history.1 mark
   5. 5.(v) Two search engines: Google and Bing. OR (vi) (c) To assess how well the chosen model will work in future.1 mark

   Marks split from a structure derived from past papers (checked twice); not CBSE’s official scheme

2. Q2Case-based4 marksAsked: 2026

   Read the following paragraph and answer the questions that follow : PQR Security Solutions has designed an AI Model to detect cyber attacks on E-Commerce websites. For this, various network activities were monitored and analyzed on one of the websites. The model was tested on a dataset of 1500 network activities. Out of these, the model correctly predicted that 1000 were cyber attacks. It also correctly identified that 250 were not cyber attacks. However, the model predicted that 200 were cyber attacks but actually they were not. Additionally, it predicted that 50 were not cyber attacks but they actually were. (A) Draw the confusion matrix based on the given scenario. (B) How many total cases are True Negative in the above scenario ? (C) Calculate Precision.

   

   Medha Ma’am

   Ask Medha’s AI teacher

   Hear it from Medha Ma’am · 130 s

   Model answer · confusion matrix, precision

   1. 1.In the given scenario, the model correctly predicted that 1000 were cyber attacks, which is the True Positive (TP) value. It correctly identified that 250 were not cyber attacks, which is the True Negative (TN) value. It predicted that 200 were cyber attacks but actually they were not, which is the False Positive (FP) value. It predicted that 50 were not cyber attacks but they actually were, which is the False Negative (FN) value. So the confusion matrix has TP = 1000, FP = 200, FN = 50 and TN = 250.2 marks
   2. 2.The total number of True Negative cases in the given scenario is 250, since the model correctly identified that 250 network activities were not cyber attacks.1 mark
   3. 3.Here, TP = 1000 and FP = 200. Precision = TP / (TP + FP) = 1000 / (1000 + 200) = 1000 / 1200 = 0.83.1 mark

   Diagram: Draw a 2x2 confusion matrix table with predicted values as columns and actual values as rows, labelled Predicted Attack / Predicted No Attack across the top and Actual Attack / Actual No Attack down the side; fill the cells with TP = 1000, FN = 50, FP = 200, TN = 250.

   • Label each cell of the matrix clearly as TP, TN, FP, FN with its value.
   • State the value 250 again as the answer for True Negative cases in point 2.
   • Write the Precision formula TP/(TP+FP) before substituting values and give the final answer as 0.83.

   Marks split from a structure derived from past papers (checked twice); not CBSE’s official scheme

3. Q3Short answer4 marksAsked: 2025

   An AI model has been developed to test samples (blood / urine / sputum etc.) collected for the diagnosis of diseases (diabetes, liver infection etc.). The model was tested on a data set of about 630 tests and the resulting Confusion Matrix is as follows : (A) How many cases in the above information are True Negative ? (B) Calculate Precision, Recall and F1 Score.

   

   Medha Ma’am

   Ask Medha’s AI teacher

   Hear it from Medha Ma’am · 110 s

   Model answer · confusion matrix, precision, recall, F1 score

   1. 1.The value of True Negative (TN) is 300. True Negatives(TN) are the cases that were correctly diagnosed as not having the disease. From the given Confusion Matrix, the number of True Negative cases is 300.1 mark
   2. 2.Precision is the ratio of the total number of correctly classified positive examples and the total number of predicted positive examples. Precision = TP/(TP+FP). Substituting the values of TP and FP from the Confusion Matrix, Precision = 0.833.1 mark
   3. 3.The recall is the measure of our model correctly identifying True Positives. Recall = TP/(TP+FN). Substituting the values of TP and FN from the Confusion Matrix, Recall = 0.714.1 mark
   4. 4.F1 Score is calculated using the formula F1 Score = 2 * (Precision * Recall)/(Precision + Recall). Substituting the calculated values of Precision = 0.833 and Recall = 0.714, F1 Score = 0.769.1 mark

   • Write the value 300 clearly as True Negative (TN) with its definition.
   • Show the formula TP/(TP+FP) before substituting values for Precision, and TP/(TP+FN) for Recall.
   • Write the F1 Score formula 2*(Precision*Recall)/(Precision+Recall) and use the already calculated Precision and Recall values to get 0.769.

   Marks split from a structure derived from past papers (checked twice); not CBSE’s official scheme

4. Q4Short answer4 marksAsked: 2024

   A binary classification model has been developed to classify news articles as either "Fake News" or "Real News". The model was tested on a dataset of 500 news articles, and the resulting confusion matrix is as follows : Confusion Matrix Reality Yes No Predicted Yes 45 15 No 20 420 (A) How many total cases are True Negative in the above scenario ? (B) Calculate Precision, Recall and F1-Score.

   How the model answer begins

   1. The total number of True Negative cases in this confusion matrix is 420. True Negative(TN) is the outcome of the model correctly predicting the negative class, that is, the cases where the predicted value is No and the…

   The full answer from Medha Ma’am →
5. Q5Long answer4 marksAsked: 2023

   Recently the country was shaken up by a series of earthquakes which has done a huge damage to the people as well as the infrastructure. To address this issue, an AI model has been created which can predict if there is a chance of earthquake or not. The confusion matrix for the same is : [Figure] (i) How many total cases are True Negative in the above scenario ? (ii) Calculate precision, recall and F1 score.

   How the model answer begins

   1. The total number of True Negative (TN) cases in the given confusion matrix is 90. True Negatives (TN) are the cases that were correctly predicted as not having a chance of earthquake, while the False Positives (FP) in…

   The full answer from Medha Ma’am →
6. Q6Very short answer2 marksAsked: 2026

   Explain Train-test split technique with respect to machine learning algorithm.

   How the model answer begins

   1. The train-test split is a technique for evaluating the performance of a machine learning algorithm.

   The full answer from Medha Ma’am →
7. Q7Short answer2 marksAsked: 2025

   Suppose you are developing an AI model to detect fraudulent financial transaction risk. Describe False Positives and False Negatives in this context.

   How the model answer begins

   1. False Positive (FP) is the outcome of the model wrongly predicting the negative class as positive class. In this scenario of detecting fraudulent financial transaction risk, a legitimate transaction is incorrectly…

   The full answer from Medha Ma’am →
8. Q8Short answer2 marksAsked: 2024

   13. What do you mean by Evaluation of an AI model ? Also explain the concept of overfitting with respect to AI model Evaluation.

   How the model answer begins

   1. Evaluation of an AI model is the process of using different evaluation metrics to understand a machine learning model's performance. It is the process of measuring the performance of the model using various parameters…

   The full answer from Medha Ma’am →
9. Q9Short answer2 marksAsked: 2023

   Define Confusion Matrix.

   How the model answer begins

   1. A confusion matrix is a handy presentation of the accuracy of a model with two or more classes, used to evaluate the performance of a classification model.

   The full answer from Medha Ma’am →
10. Q10Very short answer1 markAsked: 2023

    Two conditions when prediction matches with the reality are true positive and ________.

    How the model answer begins

    1. True Negative is the condition when the prediction matches with the reality, that is, True Negative (TN) is the outcome of the model correctly predicting the negative class.

    The full answer from Medha Ma’am →
11. Q11Very short answer1 markAsked: 2023

    With reference to AI domain, expand the term CV.

    How the model answer begins

    1. With reference to the AI (Artificial Intelligence) domain, the full form of CV is Computer Vision. In this abbreviation, the letter C stands for Computer and the letter V stands for Vision. So the term CV expands to…

    The full answer from Medha Ma’am →
12. Q12Very short answer1 markAsked: 2023

    Under ________, one looks at various parameters which affect the problem we wish to solve, as this would make many lives better.

    How the model answer begins

    1. The process of looking at various parameters which affect the problem we wish to solve, as this would make many lives better, is called 'Problem Scoping'.

    The full answer from Medha Ma’am →
13. Q13Very short answer1 markAsked: 2023

    In this learning model, the data set which is fed to the machine is labelled. Name the model.

    How the model answer begins

    1. The name of this learning model, in which the data set fed to the machine is labelled, is Supervised Learning.

    The full answer from Medha Ma’am →
14. Q14Very short answer1 markAsked: 2026

    State True or False : In machine learning, the error is used to see how accurately the model can predict data.

    How the model answer begins

    1. True. In Machine Learning, the error is used to see how accurately our model can predict data it uses to learn new, unseen data.

    The full answer from Medha Ma’am →

Objective questions with the board’s key (10)

1. 1. CBSE 2026

   In the context of autonomous vehicle safety systems, which type of error would be most critical to minimize ?

   • (A) False Positive (detecting danger when there isn't any)
   • (B) False Negative (failing to detect actual danger) ✓
   • (C) True Positive (detecting danger correctly)
   • (D) True Negative (correctly identifying that there is no danger)

   Answer: (B)

2. 2. CBSE 2026

   What is the range of possible pixel values in a byte image format ?

   • (A) 0 to 100
   • (B) 0 to 255 ✓
   • (C) 1 to 256
   • (D) -128 to 127

   Answer: (B)

3. 3. CBSE 2026

   An e-commerce platform analyzes customer purchase patterns to recommend “Customers who bought product x also bought product y.” This uses :

   • (A) Classification model
   • (B) Regression model
   • (C) Association model ✓
   • (D) Clustering model

   Answer: (C)

4. 4. CBSE 2026

   Since AI is actually being used as a tool for decision making/influencing, we need to ensure that AI makes ethically acceptable recommendations. Which of the following is a key factor that can knowingly or unknowingly influence our decision making while designing an AI model ?

   • (A) Intuition and values ✓
   • (B) Algorithm efficiency
   • (C) Data storage capacity
   • (D) Processing speed

   Answer: (A)

5. 5. CBSE 2026

   Which NLP application is helpful in converting natural speech to text in real time ?

   • (A) Keyword Extraction tool
   • (B) Translation of books from English to Hindi language
   • (C) Auto generated captions on YouTube ✓
   • (D) Classifying raw text into pre-defined groups

   Answer: (C)

6. 6. CBSE 2026

   In Supervised learning, what is the purpose of the testing dataset ?

   • (A) To train the model.
   • (B) To evaluate the accuracy of the model. ✓
   • (C) To create new features.
   • (D) To label the data.

   Answer: (B)

7. 7. CBSE 2026

   In a fire alarm system, if the model predicts “Fire Present” when there is actually no fire, this is classified as :

   • (A) True Positive (TP)
   • (B) True Negative (TN)
   • (C) False Positive (FP) ✓
   • (D) False Negative (FN)

   Answer: (C)

8. 8. CBSE 2026

   Assertion (A): Bioethics is an example of a value-based framework for AI. Reason (R): Bioethics deals with ethical issues related to health, medicine and biology.

   • (A) Both (A) and (R) are true and (R) is the correct explanation of (A).
   • (B) Both (A) and (R) are true, but (R) is not the correct explanation of (A).
   • (C) (A) is true, but (R) is false.
   • (D) (A) is false, but (R) is true. ✓

   Answer: (D)

9. 9. CBSE 2026

   Which scenario best represents a regression problem?

   • (A) Identifying whether an email is spam.
   • (B) Grouping customers according to behaviour.
   • (C) Predicting tomorrow's temperature. ✓
   • (D) Recognising faces in a photo.

   Answer: (C)

10. 10. CBSE 2026

    Which of the following best describes the relationship between pixels and image resolution?

    • (A) Having more pixels reduces the image quality.
    • (B) Pixels and resolution are not related concepts.
    • (C) Resolution depends only on the size of the image file.
    • (D) The number of pixels in an image is known as resolution. ✓

    Answer: (D)

Questions parents and students ask

How many marks does "Evaluating Models" carry in CBSE Class 10 Artificial Intelligence?

In the CBSE papers of 2023–2026, this chapter carried about 10 marks and about 9 questions a year; by marks it ranks 2 in Artificial Intelligence.

Which are the most important (VVI) questions of "Evaluating Models"?

The 14 questions above are from the board’s real papers, each with the years it was asked; the most-asked come first with their full model answers. All 14 written questions of this chapter, with answers, are in the Ask Medha app.

Do the answers follow the board’s marking scheme?

Yes. CBSE publishes its marking schemes; every model answer here is written point by point to the marks the board’s own scheme gives, and checked by a second, independent model.