IT-802-A Machine Learning - Jun 2025 Question Paper

Under each question, How to answer lists the points an answer should cover, outlined by CampusPrep from the syllabus. Check the details against your textbook.

1. 7 Marks
   aExplain the difference between supervised, unsupervised, semi-supervised and reinforcement learning Provide practical examples for each paradigm.
   How to answer

   • Define supervised, unsupervised, semi-supervised, reinforcement learning
   • Compare on labels, feedback, objective
   • Give one practical example for each paradigm
   • Summarize when each is used
2. 7 Marks
   bWhat are the key factors affecting the performance of a machine learning model? Discuss their significance with examples.
   How to answer

   • List key factors: data quality/quantity, features, model choice, hyperparameters, overfitting/underfitting
   • Explain significance of each factor
   • Illustrate with examples
   • Conclude on performance impact
3. 7 Marks
   aCompare finite and infinite hypothesis spaces in machine learning. What are the implications of each on model complexity and generalization?
   How to answer

   • Define finite vs infinite hypothesis spaces
   • Compare cardinality and model complexity
   • Discuss implications for overfitting and generalization
   • Give examples of each
4. 7 Marks
   bDescribe the key concepts behind Support Vector Machines (SVM) for linear classification. How is the margin optimized?
   How to answer

   • Define SVM, hyperplane, support vectors, margin
   • Explain hard-margin objective for linear classification
   • Describe margin maximization optimization formulation
   • Note support vectors determine decision boundary
5. 7 Marks
   aExplain the role of activation functions in a multilayer perceptron. How do they enable the learning of nonlinear patterns?
   How to answer

   • Define activation function in MLP
   • Explain need for non-linearity
   • Compare sigmoid, tanh, ReLU examples
   • Show how stacking enables nonlinear decision boundaries
6. 7 Marks
   bDifferentiate between linear regression, multiple linear regression, and logistic regression. Provide examples where each would be appropriate.
   How to answer

   • Define linear, multiple linear, logistic regression
   • Compare output type: continuous vs categorical and hypothesis functions
   • Give appropriate use-case example for each
   • Summarize differences in table
7. 7 Marks
   aExplain the ID3 algorithm for building decision trees. How does it use entropy and information gain?
   How to answer

   • Outline ID3 top-down greedy tree building steps
   • Define entropy and information gain formulas
   • Explain attribute selection by maximum gain
   • Note stopping criteria and limitations
8. 7 Marks
   bDiscuss the role of meta-learners in stacking and how they contribute to the overall model performance.
   How to answer

   • Define stacking and meta-learner concept
   • Explain training on base-learner predictions
   • Discuss how it combines strengths and reduces error
   • Note risk of overfitting and cross-validation
9. 7 Marks
   aWhat is the bagging technique in ensemble learning, and how does it reduce variance in predictions?
   How to answer

   • Define bagging and bootstrap sampling
   • Explain parallel training and aggregation
   • Show variance reduction reasoning
   • Give Random Forest as example
10. 7 Marks
    bCompare and contrast the different model combination schemes used in ensemble learning.
    How to answer

    • List combination schemes: voting, averaging, weighted, stacking, boosting
    • Compare principles and weighting
    • Contrast bias/variance effects
    • Note suitable scenarios
11. 7 Marks
    aCompare PCA and Locally Linear Embedding (LLE). When would you prefer LLE over PCA?
    How to answer

    • Outline PCA linear variance-based reduction vs LLE manifold preservation
    • Compare assumptions and computation
    • State when nonlinear manifold favours LLE
    • Give example datasets
12. 7 Marks
    bDescribe Gaussian Mixture Models (GMM) and how they model data clusters as probabilistic distributions. What makes GMMs suitable for overlapping clusters?
    How to answer

    • Define GMM as mixture of Gaussians with weights
    • Explain EM estimation and soft clustering
    • Show probabilistic membership for overlap
    • Contrast with hard k-means
13. 7 Marks
    aCompare AGNES (Agglomerative Nesting) and DIANA (Divisive Analysis) hierarchical clustering methods. What are their advantages and limitations?
    How to answer

    • Define AGNES bottom-up vs DIANA top-down
    • Compare procedure and dendrogram construction
    • List advantages and limitations of each
    • Note scalability and decision criteria
14. 7 Marks
    bWhat is frequent pattern mining, and why is it important in fields like market basket analysis?
    How to answer

    • Define frequent pattern and support
    • Explain market basket example
    • Discuss use in association rules and decisions
    • Note scalability importance
15. 7 Marks
    aDiscuss the process of constructing a Bayesian Belief Network, including defining nodes, edges and conditional probabilities.
    How to answer

    • Define Bayesian Belief Network DAG concept
    • Explain steps: identify variables, define structure/edges, specify CPTs
    • Discuss conditional independence
    • Give small example network
16. 7 Marks
    bWhat are the assumptions of the Naive Bayes Classifier? How do these assumptions simplify the computation of probabilities?
    How to answer

    • State conditional independence assumption formula
    • Explain simplification of joint probability
    • Show computation reduction with example
    • Note limitations of assumption