(SEM VI) THEORY EXAMINATION 2024-25 MACHINE LEARNING TECHNIQUES

Machine Learning Techniques (BCDS062)

B.Tech Semester VI – Exam-Ready Notes

SECTION A

(Attempt all | 2 × 7 = 14 marks)
Write 2–3 crisp lines for each answer.

(a) Supervised vs Unsupervised Learning

Supervised learning uses labeled data to learn a mapping from inputs to outputs (e.g., classification, regression).

Unsupervised learning works on unlabeled data to discover hidden patterns (e.g., clustering, association).

(b) Role of Hypothesis in a Learning System

A hypothesis represents a candidate function that maps inputs to outputs.
The learning process selects the best hypothesis that fits the training data.

(c) Hyperplane in SVM

A hyperplane is a decision boundary that separates data points of different classes.
SVM chooses the hyperplane with maximum margin.

(d) Linear vs Logistic Regression

(e) Inductive Bias

Inductive bias is the set of assumptions a learning algorithm uses to generalize from training data to unseen data.

(f) Gradient Descent

Gradient descent is an optimization algorithm that minimizes a loss function by iteratively updating parameters in the direction of steepest descent.

(g) Markov Decision Process (MDP)

An MDP is defined by states, actions, rewards, transition probabilities, and discount factor, used to model decision-making in reinforcement learning.

SECTION B

(Attempt any THREE | 7 × 3 = 21 marks)

(a) Naïve Bayes vs Bayesian Belief Networks

(b) Why SVM is Effective for High-Dimensional Data

SVM works well in high-dimensional spaces because it depends on support vectors, uses kernel functions, and avoids overfitting through margin maximization.
Example: text classification with thousands of features.

(c) Working of k-NN, Advantages & Limitations

k-NN classifies a point based on the majority class of its nearest neighbors.
Advantages: simple, no training phase.
Limitations: high computation cost, sensitive to noise and choice of k.

(d) CNN Architecture (Basic)

A CNN consists of:

Convolution layer – feature extraction         Pooling layer – dimensionality reduction

Activation layer – non-linearity                    Fully connected layer – classification
(Always draw a neat diagram in exam.)

(e) Genetic Algorithm (GA) Cycle

GA cycle includes:                                          Initialization

Selection                                                        Crossover

Mutation                                                        Replacement
Used for optimization and search problems.

SECTION C

(Attempt any ONE | 7 marks)

(a) Fraud Detection using Machine Learning

ML algorithms detect fraud by learning patterns of normal and abnormal behavior.
Techniques include logistic regression, decision trees, neural networks, and anomaly detection.
Real-time software: SAS Fraud Management, IBM Safer Payments, FICO Falcon.

OR

(b) Regression vs Classification vs Clustering