(SEM VII) THEORY EXAMINATION 2022-23 MACHINE LEARNING

SECTION A – Short Answers (2 Marks Each)

(a) Applications of Machine Learning in Industry 4.0

Machine Learning is used in Industry 4.0 for predictive maintenance, smart manufacturing, quality inspection, robot automation, supply chain optimization, and fault detection using real-time data.

(b) Example of classification problem

Email spam detection, where emails are classified as spam or not spam, is a classification problem.

(c) Training data vs Testing data

Training data is used to train the ML model, while testing data is used to evaluate the performance of the trained model on unseen data.

(d) Polynomial Regression

Polynomial regression models the relationship between dependent and independent variables as an nth-degree polynomial, useful when data is non-linear.

(e) Multidimensional Scaling (MDS)

MDS is a dimensionality reduction technique that represents high-dimensional data in lower dimensions while preserving distances between data points.

(f) K-Means Clustering

K-Means is an unsupervised learning algorithm that groups data into k clusters by minimizing the distance between data points and cluster centroids.

(g) Artificial Intelligence (AI)

Artificial Intelligence is the field of computer science that enables machines to think, learn, and make decisions similar to humans.

(h) Root Node vs Decision Node

Root node is the topmost node of a decision tree, while a decision node splits data based on a condition.

(i) Reproduction in Genetic Algorithm

Reproduction is the process of selecting the best individuals from a population to generate offspring for the next generation.

(j) Reinforcement learning vs Deep learning

Reinforcement learning learns through reward and punishment, while deep learning uses neural networks with multiple layers to learn patterns from data.

SECTION B – Long Answers (10 Marks Each)

(a) Supervised, Unsupervised and Reinforcement Learning

Supervised learning uses labeled data and includes classification and regression problems. The model learns by mapping inputs to known outputs.

Unsupervised learning uses unlabeled data and focuses on discovering patterns such as clustering and association rules.

Reinforcement learning is based on interaction with an environment where an agent learns optimal actions using rewards and penalties.

(b) Support Vector Machine (SVM)

SVM is a supervised learning algorithm used for classification and regression. It works by finding an optimal hyperplane that maximizes the margin between different classes.

Advantages:
High accuracy, effective in high-dimensional space, memory efficient.

Disadvantages:
High computational cost, difficult to choose kernel, not suitable for very large datasets.

(c) Applications of clustering

In marketing, clustering helps in customer segmentation.
In insurance, it is used for risk analysis and fraud detection.
In earthquake studies, clustering identifies seismic zones and earthquake patterns.

(d) Backpropagation algorithm in ANN

Backpropagation is a supervised learning algorithm used in neural networks. It works in two phases: forward pass and backward pass.
In the forward pass, input is propagated to output. In backward pass, error is calculated and weights are updated using gradient descent to minimize error.

(e) Genetic Algorithm (GA)

Genetic Algorithm is an optimization technique inspired by natural evolution. It uses selection, crossover, and mutation to find optimal solutions.

Advantages:
Handles complex problems, avoids local minima, works with large search spaces.

Applications:
Scheduling, optimization, feature selection, robotics, and machine learning.

SECTION C – Long Answers (10 Marks Each)

3(a) Applications of ML in healthcare and banking

In healthcare, ML is used for disease diagnosis, medical imaging, patient monitoring, and drug discovery.
In banking, ML is applied for fraud detection, credit scoring, customer service chatbots, and risk assessment.

3(b) ML applications in Netflix, Facebook, and Amazon

Netflix uses ML for movie recommendations.
Facebook uses ML for face recognition, friend suggestions, and news feed ranking.
Amazon uses ML for product recommendations, demand prediction, and dynamic pricing.

4(a) Regression as supervised learning & comparison with classification

Regression is a supervised learning technique because it uses labeled data to predict continuous values.
Classification predicts discrete class labels, while regression predicts numerical values.

4(b) Regression in machine learning

Regression estimates relationships between variables.
Example: Predicting house price based on area, location, and number of rooms.

5(a) K-Means clustering (Numerical – Method)

Given 8 points and 3 initial centroids, distances are calculated using Euclidean distance, points are assigned to nearest centroid, and new centroids are calculated.
The process is repeated for two iterations to form final clusters.

5(b) Nearest Neighbor clustering

Data points are grouped based on distance threshold t = 4.
Points within threshold distance form a cluster; remaining points create new clusters.

6(a) Decision tree, entropy & information gain

Entropy measures impurity:

Entropy=−∑plog⁡2pEntropy = -\sum p \log_2 pEntropy=−∑plog2​p

Information Gain is:

IG=Entropy(parent)−∑Entropy(children)IG = Entropy(parent) - \sum Entropy(children)IG=Entropy(parent)−∑Entropy(children)

Decision tree is constructed using attribute with highest information gain.

6(b) ANN forward pass error calculation

Hidden layer inputs and outputs are calculated using weights and bias.
Output layer values are computed, and total error is obtained using:

E=12∑(T−O)2E = \frac{1}{2}\sum (T - O)^2E=21​∑(T−O)2 

7(a) Genetic Algorithm with example

GA solves optimization problems by evolving solutions over generations.
Example: Finding optimal route in Travelling Salesman Problem.

7(b) Reinforcement learning and its types

Reinforcement learning allows an agent to learn through interaction with environment.
Types include positive reinforcement, negative reinforcement, model-based, and model-free learning.