SECTION-B 2.

Unsupervised Learning
[Link] in detail about the types of Machine learning with necessary diagrams Unsupervised learning is a type of machine learning where the model is trained
using unlabeled data, meaning no predefined output is given. The main goal is to
Machine Learning can be broadly classified into three main types based on the
discover hidden patterns, structures, or relationships within the data. The
nature of learning and the availability of data labels. These are Supervised
algorithm learns by analyzing similarities and differences among data points.
Learning, Unsupervised Learning, and Reinforcement Learning.
Unsupervised learning is commonly used for clustering, association, and
1. Supervised Learning
dimensionality reduction tasks
Supervised learning is a type of machine learning where the model is trained using
labeled data, meaning each input data is associated with a known output. The
objective is to learn a mapping function from inputs to outputs so that the model
can predict correct outputs for new, unseen data. During training, the model
compares its predicted output with the actual output and adjusts itself to
minimize the error. Supervised learning is mainly used for classification and
regression problems. In classification, the output is a class label (such as spam or
not spam), whereas in regression, the output is a continuous value (such as house
price prediction). Common algorithms used in supervised learning include Linear
Regression, Logistic Regression, Decision Trees, Support Vector Machines, and k-
Nearest Neighbors. 3. Reinforcement Learning
Reinforcement learning is a type of machine learning where an agent learns by
interacting with an environment. The agent takes actions, receives feedback in
the form of rewards or penalties, and learns to choose actions that maximize
cumulative reward over time. Unlike supervised learning, there are no labeled
input-output pairs.

Diagram:

Additional Classification:
Machine learning can also be categorized into Semi-Supervised Learning, which
uses a small amount of labeled data and a large amount of unlabeled data, and
Self-Supervised Learning, where the system generates its own labels from the
data. These methods are useful when labeled data is expensive or difficult to
obtain. Working Principle of Decision Tree Algorithm: The decision tree algorithm works
by recursively splitting the dataset into smaller subsets based on the most
2. Explain the Decision tree algorithm in detail informative feature.
A Decision Tree is a supervised machine learning algorithm used for classification Attribute Selection Measures: To decide the best attribute for splitting, the
and regression tasks. It represents decisions in the form of a tree-like structure, decision tree algorithm uses different measures. Information Gain is based on
where each internal node denotes a test on an attribute, each branch represents entropy and measures the reduction in uncertainty after a split.
an outcome of the test, and each leaf node represents a class label or predicted
value. The main objective of the decision tree algorithm is to create a model that Steps in Decision Tree Algorithm: First, the entire training dataset is considered as
predicts the value of a target variable by learning simple decision rules inferred the root node. Next, the best attribute is selected using an attribute selection
from training data. measure. The dataset is split into subsets based on attribute values.

Structure of a Decision Tree: A decision tree consists of a root node, which Pruning in Decision Trees: Pruning is used to reduce overfitting in decision trees.
represents the entire dataset and performs the first split. The internal nodes Pre-pruning stops tree growth early by setting limits such as maximum depth.
represent decision points based on feature values. The branches represent the Post-pruning removes unnecessary branches after the tree is fully grown.
outcomes of decisions or tests. The leaf nodes represent the final output, such as Advantages of Decision Tree Algorithm: Decision trees are easy to understand
a class label in classification or a numerical value in regression. and interpret. They can handle both numerical and categorical data.
Disadvantages of Decision Tree Algorithm: Decision trees are prone to overfitting, Major Issues in Machine Learning
especially with noisy data. Small changes in data can result in different tree
1. Data Quality and Quantity
structures.
Machine learning models depend heavily on the quality and amount of data.
3. Discuss the Major Perspectives and Issues in Machine Learning Noisy, incomplete, or biased data can lead to inaccurate models.
Machine Learning (ML) is a multidisciplinary field that can be viewed from 2. Overfitting and Underfitting
different perspectives based on how learning is defined and implemented. Along Overfitting occurs when a model learns noise in the training data, while
with these perspectives, several practical and theoretical issues influence the underfitting occurs when the model is too simple. Balancing model complexity is a
performance and reliability of ML systems. major issue in ML.
Major Perspectives in Machine Learning 3. Feature Selection and Dimensionality
High-dimensional data increases computational complexity and may degrade
1. Statistical Perspective: From the statistical perspective, machine learning is
performance due to the curse of dimensionality. Selecting relevant features is a
viewed as a process of learning patterns from data using probability and statistical
challenging issue.
inference. The focus is on estimating models that generalize well to unseen data
by minimizing error and uncertainty.
2. Computational Perspective: The computational perspective focuses on the 4. Define Machine Learning. Discuss the importance of Machine Learning with
efficiency and scalability of learning algorithms. It emphasizes algorithm design, examples
computational complexity, memory usage, and optimization techniques.
Definition of Machine Learning
3. Algorithmic or Learning Theory Perspective
Machine Learning is a branch of Artificial Intelligence that enables computer
This perspective studies the theoretical foundations of machine learning. It
systems to learn from data and improve their performance automatically without
focuses on learnability, convergence, sample complexity, and generalization
being explicitly programmed. It focuses on developing algorithms that can identify
guarantees.
patterns, make predictions, and take decisions based on past experiences.
4. Representational Perspective
Importance of Machine Learning
The representational perspective deals with how data and knowledge are
represented within a learning system. It emphasizes feature selection, feature 1. Automation of Decision-Making
extraction, and model representation such as decision trees, neural networks, and Machine learning allows systems to make intelligent decisions without human
rule-based systems. intervention. It automates complex decision-making processes by analyzing large
volumes of data.
5. Application-Oriented Perspective
This perspective focuses on applying machine learning techniques to real-world 2. Handling Large and Complex Data
problems. It emphasizes domain knowledge, problem formulation, evaluation Machine learning can process and analyze huge amounts of structured and
metrics, and deployment challenges. unstructured data efficiently. Traditional programming methods fail to handle such

data complexity. For example, social media platforms analyze user data to Environment: The environment refers to everything that the agent interacts with
understand behavior and preferences. while learning. It responds to the agent’s actions by changing its state and
providing rewards. It defines the rules, dynamics, and constraints of the learning
3. Improved Accuracy and Prediction
problem.
Machine learning models learn from historical data and continuously improve
their predictions. For example, weather forecasting systems use machine learning State: A state represents the current situation or condition of the environment as
to improve prediction accuracy. perceived by the agent. It contains all the relevant information needed by the
agent to decide its next action. For example, in a chess game, the arrangement of
4. Personalization and Recommendation
pieces on the board represents a state.
Machine learning enables personalized services by learning user preferences.
Recommendation systems suggest relevant products, movies, or music to users. Action: An action is a decision or move taken by the agent in a given state. Actions
For example, e-commerce websites recommend products based on browsing and influence the environment and cause transitions from one state to another. The
purchase history. set of all possible actions available to the agent depends on the current state.
Choosing appropriate actions is essential for achieving higher rewards.
5. Time and Cost Efficiency
Machine learning reduces manual effort and operational costs by automating Reward: A reward is a numerical feedback signal provided by the environment
repetitive tasks. It also speeds up processes that require real-time analysis. For after the agent takes an action. It indicates how good or bad the action was with
example, chatbots handle customer queries without human support. respect to the learning goal. The agent aims to maximize the total accumulated
reward.
6. Wide Range of Real-World Applications
Machine learning is widely used in healthcare for disease diagnosis, in finance for SECTION-C
risk assessment, in transportation for self-driving cars, and in education for
1. Simulate logical AND gate using neural network (perceptron). Assume w1=1,
adaptive learning systems. These applications improve efficiency and service
w2=1, b=-1 and learning rate=0.1. In case of error, apply the Hebbian learning
quality.
rule of adjusting weights and bias. Solve for two iterations. Assume that step
5. Define the terms: Agent, Environment, State, Action, and Reward in activation function has been used in the neurons.
Reinforcement Learning
Simulate Logical AND Gate using Neural Network (Perceptron)
Reinforcement Learning (RL) is a learning paradigm in which an intelligent system
learns to make decisions by interacting with its surroundings. The core
components of reinforcement learning include agent, environment, state, action,
and reward, each playing a crucial role in the learning process.
Agent: The agent is the learner or decision-maker in a reinforcement learning
system. It observes the current state of the environment and selects actions based
on a learning policy.. Examples of agents include a robot, a software program, or a
game-playing system.
The perceptron successfully simulates the AND logic gate after two iterations.
Using the step activation function and Hebbian learning rule, the network adjusts
its weights and bias to correctly classify all input patterns. This demonstrates that
a single-layer perceptron can solve linearly separable problems like the AND gate.

2. Write down the algorithm for the K-means Clustering technique. What are the
Distance Metrics used for quantitative and qualitative attributes?
K-means Clustering Algorithm and Distance Metrics
K-means Clustering Technique
K-means is an unsupervised learning algorithm used for partitioning a dataset
into K distinct clusters, where each data point belongs to the cluster with the
nearest mean.
Algorithm for K-means Clustering:Step 1: Choose the number of clusters (K)
Decide the number of clusters K to be formed in the dataset based on the
problem requirement or using methods such as the elbow method.
Step 2: Initialize cluster centroids Introduction to Naïve Bayes Classifier
Randomly select K data points from the dataset as the initial centroids, or initialize
The Naïve Bayes classifier is a probabilistic supervised learning algorithm based
them using a heuristic method.
on Bayes’ Theorem. It is widely used for classification tasks, especially when
Step 3: Assign data points to nearest centroid dealing with high-dimensional data. The term naïve refers to the assumption that
Calculate the distance between each data point and all centroids using a suitable all features are conditionally independent given the class label.
distance metric. Assign each data point to the cluster with the minimum distance.
Step 4: Update centroids
Recalculate the centroid of each cluster by computing the mean of all data points
assigned to that cluster.
Step 5: Repeat assignment and update steps
Repeat steps 3 and 4 iteratively until there is no change in cluster assignments or
the centroids converge.
Step 6: Termination
Stop the algorithm when convergence is achieved or when a maximum number of
iterations is reached.
Distance Metrics for Quantitative Attributes: For quantitative (numerical)
attributes, distance metrics measure the numerical difference between data Assumption of Naïve Bayes
points. Naïve Bayes assumes that all features are independent of each other given the
• Euclidean Distance: Measures straight-line distance and is most commonly class label. This simplifies computation and allows the likelihood to be expressed
used in K-means. as the product of individual probabilities

These metrics are suitable when data attributes are continuous and numeric. Working of Naïve Bayes Classifier

Distance Metrics for Qualitative Attributes: For qualitative (categorical) 1. First, the algorithm calculates the prior probability of each class from the
attributes, numerical distance measures are not directly applicable. training data.

• Hamming Distance: Counts the number of mismatched attributes between 2. Then, it computes the conditional probability of each feature value given
two data points. the class.
3. For a new input instance, the classifier multiplies the prior probability with
all conditional probabilities.
3. Explain the working of the Naïve Bayes Classifier and provide an example of
its application. 4. The posterior probability is calculated for each class.

5. The class with the highest posterior probability is assigned to the input 1. States (S):
instance. States represent all possible situations in which an agent can exist. Each
state contains sufficient information to make a decision.
Types of Naïve Bayes Classifiers
2. Actions (A):
• Gaussian Naïve Bayes: Used for continuous data assuming normal
Actions are the possible moves or decisions the agent can take in a given
distribution.
state. The set of available actions may vary from state to state.
Example Application: Email Spam Detection
3. Transition Probability (P):
Naïve Bayes is widely used in email spam filtering. In this application, emails are Transition probability defines the likelihood of moving from one state to
classified as spam or not spam based on the presence of certain words. The another after taking an action.
classifier calculates the probability of an email being spam given the words it
4. Reward Function (R):
contains.
The reward function provides a numerical feedback to the agent after
Advantages of Naïve Bayes performing an action in a state. It indicates how desirable a state or action
Naïve Bayes is simple and fast to implement. It works well with large datasets and is.
high-dimensional data. 5. Policy (π):
A policy defines the strategy of the agent by specifying the action to be
taken in each state. The goal of reinforcement learning is to find an optimal
4. Explain the Concept of Markov Decision Process (MDP) in Reinforcement
policy.
Learning
Working of MDP
A Markov Decision Process (MDP) is a mathematical framework used in
reinforcement learning to model decision-making problems where outcomes are In an MDP, the agent starts in an initial state and selects an action according to its
partly random and partly under the control of an agent. policy. The environment responds by transitioning to a new state and providing a
reward.
Markov Property
Objective of MDP
MDP is based on the Markov property, which states that the future state of the
system depends only on the current state and the action taken, and not on past The main objective of an MDP is to determine an optimal policy that maximizes
states. the long-term reward, often expressed as a discounted sum of future rewards.
This helps the agent make better decisions in uncertain environments.
Components of an MDP
Applications of MDP
An MDP is defined using five main components:
MDPs are widely used in robotics, game playing, autonomous vehicles, resource 4. Deep Learning
allocation, and control systems. They form the foundation of many reinforcement Deep learning is a major trend for handling complex and unstructured IoT data.
learning algorithms such as Q-learning and Policy Iteration. CNNs and LSTMs are used for image, video, and time-series sensor data.
Examples: Smart surveillance, predictive maintenance.
SECTION-D 5. Edge and Federated Learning
Edge learning processes data near IoT devices to reduce latency, while federated
1. Explain the recent trends in various learning techniques of machine learning
learning improves privacy by training models without sharing raw data.
and classification methods for IOT applications. Give the examples of various
models for IOT applications. Examples: Wearable health monitoring, smart meters.
The rapid expansion of Internet of Things (IoT) systems has led to the generation Classification Methods for IoT Applications
of large volumes of heterogeneous sensor data. Machine Learning (ML) plays a
Common classification methods used in IoT include Decision Trees and Random
vital role in extracting meaningful insights from this data and enabling intelligent,
Forest for interpretability, SVM for high-dimensional data, Naïve Bayes for
automated IoT applications.
lightweight classification, and Neural Networks for complex data patterns.
Recent Trends in Machine Learning Techniques for IoT
Examples of ML Models for IoT Applications
1. Supervised Learning
• Smart Homes – Decision Trees, Reinforcement Learning
Supervised learning is widely used in IoT applications where labeled data is
available. Algorithms such as Decision Trees, SVM, k-NN, and Naïve Bayes are used • Smart Healthcare – SVM, CNN, LSTM
for prediction and classification. 2. Explain feature representation learning and its importance in deep learning.
Examples: Smart healthcare diagnosis, fault detection in industrial IoT. Explain the need for scalable machine learning.

2. Unsupervised Learning Feature Representation Learning:

Unsupervised learning is increasingly used due to the lack of labeled IoT data. Feature representation learning refers to the process in which a machine learning
Clustering and anomaly detection techniques like K-means, DBSCAN, and PCA help model automatically learns useful features or representations from raw input
discover hidden patterns and abnormal behavior in sensor data. data instead of relying on manually engineered features
Examples: Intrusion detection, anomaly detection in smart grids. In deep learning, lower layers learn simple features such as edges or patterns,
3. Reinforcement Learning (RL) while higher layers learn more complex and abstract features. This hierarchical
Reinforcement learning is used for adaptive decision-making in dynamic IoT feature learning enables the model to capture complex relationships in data.
environments. Importance of Feature Representation Learning in Deep Learning
Examples: Smart traffic control, energy management in smart buildings.

Feature representation learning is important because it reduces the need for Another important role of a policy is managing the exploration–exploitation
manual feature engineering, saving time and effort. It improves model trade-off.
performance by learning optimal features directly from data
The policy also interacts closely with other RL components such as the value
Need for Scalable Machine Learning function and reward function. Value-based methods estimate the value of states
or actions to improve the policy, while policy-based methods directly optimize the
Scalable machine learning refers to the ability of learning algorithms to handle
policy. Thus, learning in reinforcement learning is essentially the process of policy
large-scale data efficiently in terms of computation, memory, and time.
improvement through repeated interaction with the environment.
Scalability is required to support distributed computing, parallel processing, and
Example: Robot Navigation / Game Playing
real-time learning. It enables training of complex models on large datasets using
frameworks such as Hadoop, Spark, and distributed deep learning platforms. Robot Navigation Example:
Scalable machine learning ensures faster training, better accuracy, and practical Consider a robot navigating a maze to reach a destination. The states represent
deployment in real-world applications. the robot’s positions, and actions include moving forward, turning left, or turning
right. The policy decides which movement the robot should take at each position.
Through trial and error, the robot learns a policy that avoids obstacles and reaches
3. What is the role of a policy in reinforcement learning the goal with minimum steps and maximum reward.
Role of a Policy in Reinforcement Learning: Game Playing Example:
In reinforcement learning (RL), a policy is a fundamental component that defines In a game such as chess or video games, the policy determines the moves chosen
the behavior of an agent. It specifies the rule or strategy by which an agent by the agent based on the current game state. By learning from rewards such as
selects an action when it is in a particular state of the environment. winning or losing, the agent improves its policy over time to make better strategic
decisions and outperform opponents.
The primary role of a policy is to guide the agent’s decision-making process so as
to maximize the cumulative reward over time. Every action chosen by the agent
affects future states and rewards; hence, the quality of the policy directly impacts
the learning performance
Policies can be classified into deterministic and stochastic policies. In a
deterministic policy, the agent selects a fixed action for a given state, making the
behavior predictable. In contrast, a stochastic policy assigns probabilities to
possible actions, allowing the agent to choose different actions in the same state.
Stochastic policies are useful in complex environments where exploration is
important.