(SEM VII) THEORY EXAMINATION 2023-24 MACHINE LEARNING

KME074 – MACHINE LEARNING

B.Tech (SEM VII) – Theory Examination
Time: 3 Hours | Max Marks: 100

SECTION A

(Attempt all questions in brief – 2 × 10 = 20 marks)

(a) Define Machine Learning and explain its significance.

Machine Learning (ML) is a branch of artificial intelligence that enables systems to learn patterns from data and improve performance without explicit programming. Its significance lies in automation, prediction accuracy, decision-making, and handling large-scale data in modern technology.

(b) Differentiate between Artificial Intelligence (AI) and Machine Learning (ML).

Artificial Intelligence is a broader concept that aims to create intelligent systems that mimic human intelligence, while Machine Learning is a subset of AI that focuses on learning from data and improving performance automatically.

(c) Difference between classification and regression in supervised learning.

Classification predicts discrete class labels (e.g., spam or not spam), whereas regression predicts continuous numerical values (e.g., house price or temperature).

(d) Types of Support Vector kernels.

Common support vector kernels are:                       Linear kernel

Polynomial kernel                                                     Radial Basis Function (RBF) kernel

Sigmoid kernel

(e) Explain Multidimensional Scaling (MDS).

Multidimensional Scaling is a dimensionality reduction technique that represents high-dimensional data in lower dimensions while preserving the distance or similarity between data points.

(f) How does K-Means clustering work?

K-Means clustering divides data into K clusters by assigning points to the nearest centroid and iteratively updating centroids until convergence.

(g) Define Backpropagation Algorithm.

Backpropagation is a learning algorithm used in neural networks to minimize error by adjusting weights using gradient descent and error propagation from output to input layers.

(h) Basics of Decision Tree algorithm.

Decision Tree is a supervised learning algorithm that splits data into subsets based on feature values using criteria such as information gain or Gini index.

(i) Meaning of reproduction in Genetic Algorithm.

Reproduction is the process of selecting the best individuals from a population to generate offspring for the next generation, ensuring survival of fittest solutions.

(j) Difference between reinforcement learning and deep learning.

Reinforcement learning focuses on learning optimal actions through rewards and penalties, while deep learning uses deep neural networks to learn representations from large datasets.

SECTION B

(Attempt any three – answers provided for ALL)

2(a) Fundamental concepts of Machine Learning and applications in Mechanical Engineering

Machine Learning involves concepts such as data collection, feature extraction, model training, testing, and evaluation. It allows systems to recognize patterns, make predictions, and automate decisions.

Significance:
ML improves efficiency, accuracy, and adaptability in complex systems.

Applications in Mechanical Engineering:

Predictive maintenance of machines                                   Fault detection in manufacturing systems

Optimization of production processes                                Robotics and automation

2(b) Bias and Variance in Machine Learning

Bias refers to errors due to overly simplistic models, leading to underfitting.
Variance refers to errors due to overly complex models, leading to overfitting.

High bias results in poor training performance, while high variance results in poor generalization.

Balancing Strategies:                                                        Cross-validation

Regularization                                                                     Increasing training data

Model complexity tuning

2(c) Unsupervised Learning, K-Means and EM Algorithm

Unsupervised learning finds patterns in unlabeled data.

K-Means Clustering:
Groups data into K clusters based on distance from centroids.

Expectation-Maximization (EM):
Iterative algorithm that estimates parameters using probability distributions.

Applications:

Customer segmentation                                    Image compression

Market analysis

2(d) Decision Trees and ID3 Algorithm

Decision Trees split data using features that provide maximum information gain.

ID3 Algorithm:                                                Uses entropy and information gain

Builds tree top-down                                        Selects best attribute at each step

Challenges:                                                     Overfitting

Handling continuous data                               Bias toward multi-valued attributes

Solutions: Pruning, ensemble methods, feature selection.

2(e) Genetic Algorithm with example, advantages and applications

Genetic Algorithm (GA) is an optimization technique inspired by natural evolution.

Example:
Optimizing machine scheduling by selecting best task sequences.

Advantages:                                                          Global search capability

Works well for complex problems                          Does not require gradient information

Applications:                                                         Optimization problems

Robotics                                                                  Engineering design

SECTION C

3(a) Components of a Machine Learning System

Key components include:                                      Data collection and preprocessing

Feature selection                                                   Model selection

Training and testing                                              Evaluation metrics

Challenges: Data quality, overfitting, scalability, and computational cost.

3(b) Difference between Data Science and Machine Learning

Data Science focuses on extracting insights from data using statistics, visualization, and ML, whereas ML focuses on building models that learn automatically.

Overlap: Data analysis and prediction               Difference: Data Science is broader; ML is model-centric.

4(a) Support Vector Machines (SVM) and case study

SVM is a supervised learning algorithm that finds an optimal hyperplane separating data classes.

Kernels: Linear, Polynomial, RBF, Sigmoid        Challenges: Kernel selection, scalability

Case Study – Car Price Prediction:
SVM regression uses features like mileage, age, engine capacity to predict car prices accurately.

4(b) Regression in Machine Learning

Regression predicts continuous values.

Examples:

Linear regression for salary prediction Polynomial regression for curve fitting

Multiple regression for house price prediction

5(a) K-Means clustering numerical problem (conceptual answer)

Using given initial centroids, distances are calculated using Euclidean distance, points are assigned to nearest clusters, centroids are updated, and the process is repeated until convergence (two iterations as required).

5(b) Multidimensional Scaling (MDS) and Linear Discriminant Analysis (LDA)

MDS reduces dimensionality while preserving distances.
LDA maximizes class separability.

Both are used in pattern recognition and data visualization.

6(a) Neural Networks, Perceptron and Backpropagation

A neural network consists of interconnected neurons.

Perceptron: Single-layer linear classifier
Backpropagation: Multi-layer training algorithm using gradient descent

Universal Approximation Theorem:
Neural networks can approximate any continuous function.

6(b) Convolutional Neural Networks (CNNs)

CNNs are specialized neural networks for image and signal processing.

Layers:                                                               Convolution layer

Pooling layer                                                      Fully connected layer

Case Study:
CNNs in self-driving cars detect lanes, pedestrians, and traffic signs.

7(a) Genetic Algorithm – explanation and advantages

GA evolves solutions using selection, crossover, and mutation.

Advantages:                                                         Robust optimization

Handles complex search spaces                            Parallel processing capability

7(b) Reinforcement Learning (RL)                    RL trains agents through rewards and penalties.

Comparison:                                                        Supervised learning uses labeled data

Unsupervised learning finds patterns                   Reinforcement learning learns via interaction

Applications:                                                       Game playing (AlphaGo)

Robotics                                                                Autonomous vehicles