Machine Learning Exam Solutions
Q1a
Apply K-Nearest Neighbor Algorithm (KNN) on the following data. Predict the student result for
values Physics = 6 marks, Chemistry = 8 marks.
Consider number of neighbors K = 3 and Euclidean Distance as distance measure.
Dataset:
| Physics (marks) | Chemistry (marks) | Result |
|------------------|--------------------|--------|
|4 |3 | Fail |
|6 |7 | Pass |
|7 |8 | Pass |
|5 |5 | Fail |
|8 |8 | Pass |
Solution:
Calculate Euclidean distances, sort, and predict using majority vote. Result: **Pass**
Q1b
Explain Support Vector Machine (SVM) classification algorithm with a suitable example.
Support Vector Machine (SVM) works by finding the hyperplane that maximally separates data
points of different classes. Key concepts include hyperplane, margin, and support vectors. For
non-linearly separable data, kernels (e.g., RBF) are used to transform data to a higher-dimensional
space.
Q2a
Explain any 4 evaluation measures of binary classification with examples.
1. Accuracy: Ratio of correct predictions. Example: 90% for 90 correct out of 100.
2. Precision: Ratio of true positives to total positive predictions. Example: 80% if 8 of 10 are
correct.
3. Recall: Ratio of true positives to all actual positives. Example: 80% if 8 out of 10 actual
positives are found.
4. F1-Score: Harmonic mean of Precision and Recall. Example: 74% for Precision = 80%, Recall
= 70%.
Q2b
Explain construction of a multi-classifier using One-vs-All and One-vs-One approaches.
- One-vs-All (OvA): Train N classifiers, each separating one class from others.
- One-vs-One (OvO): Train N(N-1)/2 classifiers for every pair of classes.
Q3a
Explain K-Means clustering algorithm, its advantages and disadvantages.
K-Means partitions data into K clusters using distance from centroids. Advantages include
simplicity and efficiency. Disadvantages include sensitivity to initialization and struggles with
non-spherical clusters.
Q4a
Elaborate need of clustering and explain how the elbow method is used to decide the value of
cluster k.
Clustering groups similar data. The elbow method plots Within-Cluster-Sum-of-Squares (WCSS)
vs. k. The "elbow point" is the optimal k.
Q4b
Explain Divisive Hierarchical Clustering (DHC) algorithm with example.
DHC starts with all data as one cluster and splits recursively. Example: Dividing a dataset based
on maximum dissimilarity until each point forms its own cluster.
Q5a
Differentiate the Bagging and Boosting approaches in ensemble learning.
- Bagging: Reduces variance by training multiple independent models. Example: Random Forest.
- Boosting: Reduces bias by focusing on hard-to-classify points. Example: AdaBoost.
Q5c
Explain AdaBoost algorithm in detail.
AdaBoost combines weak learners, weighting them based on accuracy. It iteratively improves by
focusing on misclassified points and adjusts model weights.
Q6c
Explain Random Forest ensembles with an example.
Random Forest uses Bagging with Decision Trees, introducing randomness in feature selection.
Example: Building trees with random feature subsets and averaging their predictions.
Q7a
Explain the following terms:
1. Markov Property: Future state depends only on the current state.
2. Bellman Equation: Recursive formula for optimal state value.
3. Markov Reward Process: Combines Markov chain with reward functions.
4. Markov Chain: State transition model depending only on the current state.
Q7b
Explain Q-Learning algorithm with example.
Q-Learning is a model-free RL algorithm. It updates Q-values iteratively to find the optimal policy.
Example: A robot navigating a grid to maximize rewards learns the shortest path.
Q8a
What is Reinforcement Learning? Explain real-time applications.
RL is a learning method where an agent interacts with an environment to maximize cumulative
rewards. Applications include robotics, gaming, autonomous vehicles, and healthcare.
Q8b
Explain the following terms:
1. Supervised Learning: Learning from labeled data. Example: Predicting house prices.
2. Unsupervised Learning: Discovering patterns in unlabeled data. Example: Customer
segmentation.
3. Reinforcement Learning: Learning via environment interactions to maximize rewards. Example:
Autonomous robots.
