Attempt PFRDA Grade A 2026 AI/ML Practice Quiz and Download PDF

1. Which statement best describes Artificial Intelligence (AI)?

It is the broad field of building systems that perform tasks requiring humanlike intelligence.
It is only the use of spreadsheets for statistical analysis.
It refers only to robots that physically resemble humans.
It is a synonym for database management.
It is limited to rule-based expert systems and excludes learning systems.

2. Which relationship among AI, Machine Learning and Deep Learning is most accurate?

AI is a subset of Deep Learning, which is a subset of Machine Learning.
Deep Learning is a subset of Machine Learning, which is a subset of AI.
Machine Learning and AI are unrelated disciplines.
Deep Learning contains all of AI and all of statistics.
Machine Learning is a subset of databases rather than AI.

3. In traditional rule-based programming, how are outputs primarily produced?

The system learns all rules automatically from labelled examples.
The model always uses reinforcement signals instead of rules.
Programmers explicitly specify rules that transform inputs into outputs.
The system cannot accept numerical inputs.
Outputs are generated only by large language models.

4. In machine learning, what is a feature?

The final predicted value produced by the model.
A legal restriction placed on model deployment.
A metric used only for regression evaluation.
An input variable or measurable attribute used by a model to make a prediction.
A fixed hardware component in an AI server.

5. What is meant by inference in a machine-learning system?

Deleting the training dataset after model creation.
Changing labels in the training data manually.
Calculating only the training loss before learning starts.
Replacing all features with random values.
Using a trained model to generate predictions on new data.

6. Which of the following is primarily a regression task?

Predicting the expected rupee value of a customer’s monthly expenditure.
Classifying an email as spam or not spam.
Assigning a transaction to one of five fraud categories.
Grouping customers into segments without labels.
Detecting whether a face matches a known identity.

7. Which of the following is primarily a classification task?

Estimating the exact market price of a bond tomorrow.
Predicting whether a loan applicant will default or not default.
Forecasting the number of transactions in the next hour.
Estimating a customer’s annual income as a continuous value.
Calculating the expected loss amount in rupees.

8. What is a model parameter in machine learning?

A user-selected setting such as the learning rate.
A label that is manually attached to every prediction after deployment.
A value learned from training data, such as a regression coefficient or neural-network weight.
A data-cleaning rule that must never change.
A performance metric chosen after testing.

9. What is a hyperparameter?

A model coefficient estimated directly from the training examples.
The true target value in a labelled dataset.
A mandatory legal disclosure for every AI model.
A configuration value selected for the learning process, such as tree depth or learning rate.
A test-set prediction that has been manually corrected.

10. What does model generalization refer to?

The ability to memorize every training example exactly.
The ability to increase the number of features without limit.
The ability to avoid using any validation data.
The ability to generate only deterministic outputs.
The ability of a model to perform well on new, unseen data.

11. What is the defining characteristic of supervised learning?

The training data contain input examples paired with known target labels or values.
The training data contain no target information at all.
The agent learns only from rewards produced by an environment.
The model is never evaluated against known outcomes.
The system can operate only on text data.

12. Which algorithm is most directly associated with predicting a continuous target using a linear relationship?

K-Means Clustering
Linear Regression
Apriori Association Mining
DBSCAN
Principal Component Analysis

13. Despite its name, Logistic Regression is commonly used for which task?

Unsupervised clustering
Dimensionality reduction
Classification
Association-rule mining
Image compression only

14. What is a decision tree designed to do?

Reduce every dataset to exactly two dimensions.
Generate only random predictions to prevent bias.
Store records in a relational database hierarchy.
Split data through a sequence of feature-based decision rules to make predictions.
Calculate the mean of the target and ignore all inputs.

15. A Random Forest primarily improves upon a single decision tree by:

Using only one deep tree with no randomness.
Removing all features except the target.
Replacing training data with test data.
Guaranteeing zero prediction error.
Combining predictions from many diverse decision trees.

16. What is the basic idea behind K-Nearest Neighbours (KNN)?

Predict using the labels or values of nearby training examples in feature space.
Always fit a straight line through the entire dataset.
Create random clusters without measuring similarity.
Use only a single global mean for all predictions.
Train a language model on token sequences.

17. Naive Bayes classifiers rely on which simplifying assumption?

All features must be perfectly correlated.
Features are conditionally independent given the class.
Every class must contain the same number of observations.
The dataset must be unlabelled.
No probability calculations are permitted.

18. What is the key objective of a Support Vector Machine classifier?

Minimize the number of rows in the dataset.
Create clusters by minimizing within-cluster variance only.
Find a decision boundary that maximizes the margin between classes.
Predict exclusively through majority class frequency.
Transform every feature into text tokens.

19. Which method is an ensemble technique that builds learners sequentially to correct earlier errors?

K-Means
PCA
Ordinary least squares without iteration
Gradient Boosting
Simple random sampling

20. A bank has historical customer data with a known label ‘churned’ or ‘not churned’. Which learning setup is most appropriate?

Unsupervised clustering
Reinforcement learning without labels
Dimensionality reduction only
Association mining only
Supervised classification