ADVANCED CONCEPTS OF MODELLING IN AI –

COMPREHENSIVE CBSE NOTES

Table of Contents
OVERVIEW

1. REVISITING AI, ML, AND DL

2. MODELLING – TYPES OF AI MODELS

3. CATEGORIES OF MACHINE LEARNING BASED MODELS

4. SUB-CATEGORIES OF SUPERVISED LEARNING

5. SUB-CATEGORIES OF UNSUPERVISED LEARNING

6. NEURAL NETWORKS

7. SUB-CATEGORIES OF DEEP LEARNING

COMPREHENSIVE DEFINITIONS SECTION
PREVIOUS YEAR QUESTIONS (PYQs) – SOLVED WITH CBSE MARKING SCHEME

OVERVIEW
This chapter covers AI modeling concepts including the differentiation between AI, ML, and DL; types
of AI models (Rule-based and Learning-based); categories of machine learning (Supervised,
Unsupervised, Reinforcement Learning); sub-categories (Classification, Regression, Clustering,
Association); and Neural Networks with their decision-making processes.

1. REVISITING AI, ML, AND DL

1.1 Differentiate between AI, ML, and DL

Artificial Intelligence (AI)

Artificial Intelligence refers to any technique that enables computers to mimic human intelligence.

An artificially intelligent machine works on algorithms and data fed to it and gives the desired
output.

AI is the umbrella terminology which covers machine learning and deep learning.

Machine Learning (ML)

Machine Learning enables machines to improve at tasks with experience.

The machine learns from new data fed to it while testing and uses it for the next iteration.

It takes into account the times when it went wrong and considers the exceptions too.
 ML is a subset of AI.

Deep Learning (DL)

Deep Learning enables software to train itself to perform tasks with vast amounts of data.
Since the system has got a huge set of data, it is able to train itself with the help of multiple
machine learning algorithms working altogether to perform a specific task.

DL is a subset of ML and the most advanced form of AI.

Relationship: Artificial Intelligence ⊃ Machine Learning ⊃ Deep Learning (Funnel approach)

1.2 Common Terminologies Used with Data

Data

Data is information in any form.

For e.g., a table with information about fruits is data.

Each row contains information about different fruits.

Features

Columns of the table are called features.

In the fruit dataset example, features may be name, color, size, etc.

Some features are special, they are called labels.

Labels

Data Labeling is the process of attaching meaning to data.

It depends on the context of the problem being solved.

For e.g., if predicting what fruit it is based on color, then color is the feature, and fruit name is the
label.
Labeled Data

Data to which some tag/label is attached.

For e.g., Name, type, number, etc.

Unlabeled Data

The raw form of data.

Data to which no tag is attached.

Training Dataset

Collection of examples given to the model to analyze and learn.

Just like a teacher teaches a topic through examples and illustrations.

A set of labeled data is used to train the AI model.

Testing Dataset
 Used to test the accuracy of the model.

Just like a class test evaluates student understanding.

Test is performed without labeled data and results are verified with labels.

2. MODELLING – TYPES OF AI MODELS

2.1 Rule-Based Approach

Definition: AI modeling where relationships or patterns in data are defined by the developer. The
machine follows rules or instructions mentioned by the developer and performs its task accordingly.

Characteristics:

Based on predefined rules and data fed to the machine

Relationship patterns defined by developers

Machine reacts accordingly to deliver desired output
Learning is static

Does not take into consideration changes in original training dataset

Example – Rule-Based Chatbot:

A clothing website has a Chatbot to answer questions about order tracking.

1. Data: Chatbot doesn't require massive dataset. Relies on predefined questions and answers.
2. Rules: Uses decision tree approach with clearly defined rules:
Rule 1: If message contains "track order," "shipment status," or "delivery," proceed to order
tracking
Rule 2: If user asks for order number, prompt them to enter it

Rule 3: If valid order number entered, retrieve tracking information

Rule 4: If invalid order number, provide error message

3. Interaction: User message analyzed based on defined rules; Chatbot responds with pre-written
answer.

Drawback: Once trained, the model cannot improvise based on feedback or adapt to new data.

2.2 Learning-Based Approach

Definition: A method where a computer learns how to do something by looking at examples or getting
feedback, similar to how we learn from experience. Instead of being explicitly programmed, the
computer learns by analyzing data and finding patterns on its own.

Characteristics:

Machine learns by itself

AI model gets trained on data fed to it
 Designs model which is adaptive to change in data

Modifies itself according to changes in data

Handles exceptions in data

Example – Learning-Based Spam Email Filter:

Provided with large dataset of emails labeled as spam or legitimate

Filter analyzes content and characteristics (words, sender info, attachments)

Uses machine learning algorithms to recognize patterns

Learns to distinguish spam from legitimate emails

Continuously adapts and improves over time

Classifies new incoming emails based on learned patterns

3. CATEGORIES OF MACHINE LEARNING BASED MODELS

3.1 Supervised Learning

Definition: In a supervised learning model, the dataset fed to the machine is labeled. The dataset is
known to the person training the machine, allowing them to label the data. A label is information used
as a tag for data.

Key Concepts:

Supervised Learning indicates having a supervisor as a teacher

Similar to a math teacher teaching using solved examples (training) then testing knowledge

Supervised Learning is when you make the machine learn by training it using labeled data

Advantages:

Clear target variable available

Can measure performance directly

Good for predictive tasks

Disadvantages:

Requires large labeled dataset

Labeling data is expensive and time-consuming

Supervised Learning Example – Currency Coins:

Problem Statement: Build a model to predict the coin based on its weight.

1 Euro weighs 5 grams

1 Dirham weighs 7 grams

1 Dollar weighs 3 grams

1 Rupee weighs 4 grams

 Feature: Weights

Label: Currency

The trained model can identify a coin based on its weight since it has already learned the patterns.

3.2 Unsupervised Learning

Definition: An unsupervised learning model works on unlabeled dataset. The data fed to the machine
is random, and the person training the model may not have information about it. These models are
used to identify relationships, patterns, and trends out of the data fed into it.

Key Concepts:

Unsupervised Learning is a type of learning without any guidance

Like a child learning to swim without supervision
The machine discovers patterns, similarities, and differences on its own

Based on unlabeled dataset

Advantages:

Works with unlabeled data

Discovers hidden patterns
No need for extensive data labeling

Disadvantages:

Hard to evaluate results

Results may not always be interpretable

Requires more computational power

Unsupervised Learning Example – Supermarket Customer Database:

Assume customer database with purchase records over period

Marketing manager wants to send grocery offer to regular grocery buyers

No predefined labels for "grocery shoppers" and "non-grocery shoppers"

Model discovers patterns and creates two clusters/groups automatically

3.3 Reinforcement Learning

Definition: This learning approach enables the computer to make a series of decisions that maximize
a reward metric for the task without human intervention and without being explicitly programmed to
achieve the task.

Key Concepts:

Machine learns through repeated trial-and-error method

Receives feedback (positive or negative) for each action

 Maximizes reward metric

No need for pre-existing knowledge or large dataset

What Makes it Different:

1. For supervised and unsupervised learning, you need good idea of data and how to solve problem

2. Frequently encounter large complex problem spaces

3. Need to respond to unforeseen environments

4. Environment may change – system needs to be adaptive

Reinforcement Learning Example – Apple Prediction:

Provide image of apple to machine; ask it to predict

Machine predicts 'cherry' → receive negative feedback (incorrect)

Machine learns it's not a cherry
Ask again with apple image

Machine predicts 'apple' → receive positive feedback (correct)

Machine learns this is an apple

Applications:

Self-driving cars (parking)

Humanoid robots (walking)
Game playing (AlphaGo)

Summary of ML Models:

Supervised learning: Used to determine relationships through training

Unsupervised learning: Used to discover new patterns from data

Reinforcement learning: Used to implement machine learning through reward mechanism

4. SUB-CATEGORIES OF SUPERVISED LEARNING

4.1 Classification Model

Definition: Data is classified according to labels. The model works on discrete dataset, meaning data
need not be continuous.

Characteristics:

Output is categorical
Discrete dataset

Assigns class labels to inputs

Examples include: hot/cold, spam/not spam, disease/no disease

Example 1 – Weather Classification:

Trained on historical weather data labeled as "hot" or "cold"

Learns patterns based on:
Location (temperature varies geographically)

Season (summer vs. winter)

High and low temperatures

Humidity

When presented with tomorrow's weather data, predicts "hot" or "cold"

Example 2 – Email Spam Classification:

Model shown tons of emails (real and spam)

Learns what makes email look like spam
Analyzes new email and decides: spam or not spam

Modern classifiers also identify: social, advertisement, notifications, etc.

Binary Classification: Two possible outcomes (e.g., yes/no, spam/not spam, disease/healthy)

Multi-class Classification: More than two possible outcomes (e.g., email categories: social,
advertisement, notifications, promotions)

4.2 Regression Model

Definition: Such models work on continuous data. They predict continuous values based on input
variables.

Characteristics:

Output is continuous value

Continuous data fed to machine

Algorithms predict continuous value based on input variables

Examples: temperature, price, income, age, etc.

Example 1 – Predicting Temperature:

Temperature is a continuous variable (can take any value within range)

Regression models well-suited for predicting continuous outputs
Example 2 – Predicting House Price:

Features/Independent Variables:

Number of bedrooms
Carpet size

Garage area

Label/Dependent Variable:
 Price

Predicts price based on parameters

Example 3 – Used Car Price Prediction:

Parameters:

Fuel type

Years of service
Number of previous owners

Kilometers driven
Transmission type (manual/automatic)

Predicts approximate selling price (continuous value)

Classification vs Regression Table:

Aspect Classification Regression

Output Type Categorical Continuous

Output Example Email spam/not spam House price in rupees

Evaluation Metric Accuracy, Confusion Matrix, Precision, Recall Mean Squared Error, R-square

Data Type Discrete Continuous

5. SUB-CATEGORIES OF UNSUPERVISED LEARNING

5.1 Clustering

Definition: In clustering, we have input data with no class labels (unlabeled data). The unsupervised
learning model divides this data into clusters based on similarity of characteristics. It differentiates data
into different clusters.

Characteristics:

Groups similar data points

Finds similarities and places in same cluster

Differentiates from objects in other clusters

No predefined classes

Unsupervised, Divide data points into different groups

Example – Jim's Music Preference:

Jim likes slow tempo and soft intensity; dislikes fast tempo and high intensity
Cluster 1: Songs with slow tempo and soft intensity (Jim likes)

Cluster 2: Songs with fast tempo and high intensity (Jim dislikes)
 If new song X has slow tempo and soft intensity, predict Jim will like it

OTT platforms (Netflix/Spotify) use similar clustering for recommendations

Difference Between Clustering and Classification:

Classification uses predefined classes; objects assigned to these

Clustering finds similarities between objects and places in same cluster

5.2 Association

Definition: Association Rule is an unsupervised learning method used to find interesting relationships
between variables from database.

Characteristics:

Finds co-occurrence patterns

Identifies relationships between items/variables
Useful for recommendations

Unsupervised, Predict output based on patterns

Example – Market Basket Analysis:

Based on purchase patterns of customers, can predict:

If Customer X buys bread, will most probably buy butter

Such associations useful for product recommendations
Stores can create bundle offers (bread + butter)

Association vs Clustering:

Aspect Association Clustering

Goal Find relationships between items Group similar items together

Method Rule-based patterns Similarity-based grouping

Example Bread + Butter recommendations Customer segment grouping

6. NEURAL NETWORKS

6.1 What is Neural Network?

Definition: Neural networks are loosely modeled after how neurons in the human brain behave. They
are a system of organizing machine learning algorithms to perform certain tasks. They can extract
data features automatically without needing programmer input.

Key Advantages:

Extract data features automatically without programmer input

Fast and efficient way to solve problems with very large datasets
 Particularly useful for image processing

Can learn complex patterns

Structure:

Divided into multiple layers

Each layer divided into blocks called nodes

Each node has its own task, passed to next layer

Neural Network consists of:

Input layer
Hidden layer(s)

Output layer

6.2 Layers of Neural Network

Input Layer:

First layer of neural network

Job is to acquire data and feed it to network
No processing occurs at input layer

Acts as data entry point

Hidden Layer(s):

Layers in which processing occurs

Name means they are hidden and not visible to user

Each node has its own machine learning algorithm
Performs computation using weights and biases on each node

Information passes from one layer to another after value from calculation passes through
activation function

Multiple hidden layers can exist depending on function complexity

Number of nodes in each layer can vary accordingly

Output Layer:

Last layer that receives processed data from final hidden layer
Similar to input layer, does not process data

Meant for user-interface

Provides final output to user
6.3 How AI Makes a Decision – The Perceptron

Example: Deciding to Go to the Park

Factors influencing decision:

Do I have a jacket?
Do I have an umbrella?

Is it sunny now?

What is the weather forecast for later?

Ranking Importance (weights can differ by person):

"Is it sunny now" – more important than "weather forecast later"

"Having a jacket" – more important than "having umbrella"

"Weather forecast for later" – least important

Perceptron Representation:

Four inputs: X1 to X4
Four weights: W1 to W3 (represent importance)
Bias: B with weight WB

Process: Sum all weighted inputs + bias, compare with threshold

Output: Decision based on threshold comparison
How Weights are Determined:

1. From experience: Personal history shows sunny day = good day

2. From personal preference: Person may weigh weather forecast more heavily
3. Bias weight (WB): Based on personal caution level

Cautious person: Higher WB (lean towards staying indoors)

Daring person: Lower WB (more likely to go out)

Example Calculation 1:

Have jacket (1), Don't have umbrella (0), Sunny now (1), Forecast rain (0)

Weights: W1=1, W2=0.5, W3=1, W4=0.8, WB=4

Calculation: (1×1) + (0×0.5) + (1×1) + (0×0.8) + (1×4) = 1 + 0 + 1 + 0 + 4 = 6

Result: 6 > threshold (0) → Decision: GO OUT

Example Calculation 2:

Don't have jacket (0), Have umbrella (1), Not sunny (0), Forecast okay (1)
Weights: W1=1, W2=0.5, W3=1, W4=0.8, WB=4

Calculation: (0×1) + (1×0.5) + (0×1) + (1×0.8) + (1×4) = 0 + 0.5 + 0 + 0.8 + 4 = 5.3

Result: 5.3 > threshold (0) → Decision: GO OUT

Key Points:

No right or wrong answers for weight values

Effects of values determine outcome (go or stay)
Everyone comes up with different decision even with same situation

This is how neural networks learn: adjusting weights based on desired output

6.4 Real-World Applications of Neural Networks

Facial recognition

Customer support chatbots

Vegetable price prediction

Image classification

Pattern recognition

7. SUB-CATEGORIES OF DEEP LEARNING

Deep Learning Models:

1. Artificial Neural Networks (ANN): Modeled on human brain and nervous system; automatically
extract features

2. Convolutional Neural Networks (CNN): Specialized for image processing; assign importance to
various aspects of image

COMPREHENSIVE DEFINITIONS SECTION

CORE AI CONCEPTS

Artificial Intelligence (AI): Any technique enabling computers to mimic human intelligence using
algorithms and data.
Machine Learning (ML): Technology that enables machines to improve at tasks through
experience and learning from data.

Deep Learning (DL): Subset of ML using multiple machine learning algorithms to train on vast
data for complex task performance.
Algorithm: Set of rules or procedures a computer follows to solve a problem or perform a task.

Model: Algorithm or process that transforms inputs into outputs to solve a given task.
DATA-RELATED DEFINITIONS

Data: Information in any form (numbers, text, images, etc.).

Feature: Measurable property (column) in dataset used as input variable.
Label: Target output or correct answer the model is to predict.

Labeled Data: Data with tags/labels attached (e.g., images labeled "cat" or "dog").

Unlabeled Data: Raw data without any tags or labels attached.

Training Dataset: Collection of labeled examples given to model to learn patterns.

Testing Dataset: Data used to evaluate model accuracy; separate from training data.
Dataset: Complete collection of data for a particular task.

MACHINE LEARNING APPROACHES

Rule-Based Approach: AI modeling with relationships defined by developers; machine follows
predefined rules.
Learning-Based Approach: Machine learns patterns from data automatically; adapts to new
data.

SUPERVISED LEARNING DEFINITIONS

Supervised Learning: Learning from labeled data with known input-output pairs; requires
supervisor.

Classification: Predicting categorical output (discrete classes like spam/not spam).

Binary Classification: Classification with two possible outcomes.
Multi-class Classification: Classification with more than two possible outcomes.
Regression: Predicting continuous numerical values (like price, temperature).

Continuous Data: Data that can take any value within a range (e.g., temperature).
Discrete Data: Data that can only take specific values (e.g., categories).

UNSUPERVISED LEARNING DEFINITIONS

Unsupervised Learning: Learning from unlabeled data; machine discovers patterns

independently.
Clustering: Grouping similar data points together without predefined classes.

Cluster: Group of similar data points formed by clustering algorithm.

Association: Finding relationships between variables/items in data.

Association Rule: Unsupervised method to find interesting relationships between database
variables.

Pattern: Recurring sequence or structure identified in data.

Similarity: How alike two data points are based on their features.
REINFORCEMENT LEARNING DEFINITIONS

Reinforcement Learning: Machine learns through trial-and-error using reward/penalty feedback.

Reward: Positive feedback given to agent for correct action.
Penalty: Negative feedback given to agent for incorrect action.

Agent: Entity that learns to make decisions based on feedback.

Environment: Context or situation agent interacts with.

NEURAL NETWORK DEFINITIONS

Neural Network: System organizing machine learning algorithms modeled after brain neurons.

Node: Individual unit in a layer that performs computations.

Layer: Set of nodes processing data together (input, hidden, output).

Input Layer: First layer; acquires and feeds data to network.

Hidden Layer: Processing layer where feature extraction and computation occur.
Output Layer: Final layer; provides result to user.
Weight: Importance factor assigned to each input in neural network node.

Bias: Additional input to neural network node to adjust output.

Activation Function: Mathematical function applied to node output to introduce non-linearity.
Perceptron: Basic unit of neural network that mimics brain neuron behavior.

Threshold: Value used to determine output decision (greater than = positive, less than =
negative).

DEEP LEARNING DEFINITIONS

Artificial Neural Networks (ANN): Deep learning model modeled on human brain; automatically
extracts features.
Convolutional Neural Network (CNN): Deep learning algorithm for image analysis; assigns
importance to image aspects.

Convolution: Mathematical operation multiplying image and kernel to extract features.

Kernel: Matrix slid across image and multiplied with input for enhanced output.

Feature Map: Output of convolution layer showing extracted features.

ReLU (Rectified Linear Unit): Function removing negative values from feature map.
Pooling Layer: Layer reducing spatial size of features while retaining important information.

Max Pooling: Returns maximum value from image portion covered by kernel.

Average Pooling: Returns average value from image portion covered by kernel.
Fully Connected Layer: Final layer using convolution/pooling results to classify image into label.

Non-linearity: Property of introducing non-linear relationships in model.

MODEL-RELATED DEFINITIONS

Overfitting: Model memorizes training data and fails on new unseen data.
Underfitting: Model is too simple to capture data patterns.
Evaluation: Process of using metrics to understand machine learning model's performance.

Accuracy: Measure of total correct predictions model makes.

Error: Difference between model's prediction and actual outcome.

Train-Test Split: Dividing dataset into training and testing subsets.

Cross Validation: Technique to evaluate model using multiple data splits.

PREVIOUS YEAR QUESTIONS (PYQs) – SOLVED WITH CBSE MARKING SCHEME

Q1. (5 Marks) Explain the Difference Between Rule-Based and Learning-Based AI

Models with Examples

Marking Scheme:

Definition of Rule-Based Model (1 mark)

Characteristics of Rule-Based Model (1 mark)

Definition of Learning-Based Model (1 mark)
Characteristics of Learning-Based Model (1 mark)
Example for each (1 mark)

Answer:

Rule-Based AI Models:

Definition: AI models where relationships and patterns in data are explicitly defined by
developers. The machine follows predefined rules and instructions.
Characteristics:

Developers define all rules

Static learning – cannot adapt to new data

Does not improve over time

Suitable for deterministic tasks with fixed patterns

Example: Order-tracking chatbot on shopping website

Uses predefined decision tree with rules like "if message contains 'track order' then proceed
to order tracking"

Returns pre-written responses based on keywords

Cannot learn from user interactions or adapt to new query types

Learning-Based AI Models:
 Definition: AI models where machine learns patterns from data automatically. The machine
designs its own algorithms based on training data.
Characteristics:
Machine learns from data independently

Adaptive – improves with new data

Continuously improves over time

Suitable for complex tasks with variable patterns

Example: Spam email filter

Trained on labeled emails (spam/legitimate)

Learns to identify patterns distinguishing spam

Continuously adapts as it encounters new emails
Improves accuracy over time

Key Difference: Rule-based is static and programmer-controlled; Learning-based is dynamic and

data-driven.

Q2. (5 Marks) Differentiate Between Supervised Learning, Unsupervised Learning,

and Reinforcement Learning. Provide One Real-World Example for Each
Marking Scheme:

Definition of Supervised Learning (1 mark)

Definition of Unsupervised Learning (1 mark)

Definition of Reinforcement Learning (1 mark)
Real-world example for each (2 marks for 3 examples, approximately)

Answer:

Supervised Learning:

Definition: Learning from labeled data where both input and expected output are provided.
Machine learns patterns to map inputs to outputs.
Key Characteristics:

Requires labeled dataset

Has clear target variable

Learns relationships through training examples

Real-World Example: Email spam classification

Train model on thousands of emails labeled "spam" or "legitimate"

Model learns patterns (specific words, sender addresses, attachments)

Used to classify new incoming emails automatically
Unsupervised Learning:

Definition: Learning from unlabeled data where no expected outputs are provided. Machine
discovers hidden patterns and relationships independently.
Key Characteristics:
Works with unlabeled data

No predefined output variable

Discovers patterns automatically

Real-World Example: Netflix recommendation system

Analyzes watch history (unlabeled) of millions of users

Identifies patterns: users who watched movie A often watch movie B

Groups similar users and recommends content accordingly

No predefined categories – system discovers patterns

Reinforcement Learning:

Definition: Learning through trial-and-error using feedback (rewards/penalties). Machine learns

what actions maximize cumulative reward.
Key Characteristics:
No labeled training data needed

Learns through feedback mechanism

Adapts to unforeseen environments
Real-World Example: Self-driving car learning to navigate
Receives reward for safe driving, reaching destination safely

Receives penalty for traffic violations, accidents

Through repeated interaction with environment, learns optimal driving decisions
Improves performance without explicit programming

Q3. (5 Marks) Explain Classification and Regression Models. Differentiate Them with
Appropriate Examples
Marking Scheme:

Definition of Classification (1 mark)

Definition of Regression (1 mark)

Characteristics comparison (1 mark)

Example for Classification (1 mark)

Example for Regression (1 mark)

Answer:
Classification Model:

Definition: Supervised learning model that predicts categorical output. Data is classified
according to predefined labels/classes.
Characteristics:
Output is categorical (discrete)

Works on discrete dataset

Assigns class labels to new data

Binary or multi-class classification

Example: Weather prediction (Hot/Cold)

Trained on historical data labeled "hot" or "cold"

Considers factors: temperature, humidity, season, location

Given tomorrow's weather parameters, predicts "hot" or "cold" category

Output is one of two categories

Regression Model:

Definition: Supervised learning model that predicts continuous numerical values. Output can take
any value within a range.
Characteristics:

Output is continuous (numerical)

Works on continuous data
Predicts specific numerical values
No fixed set of possible outputs

Example: House price prediction

Independent variables (features): number of bedrooms, area in sq ft, location
Dependent variable (label): price in rupees
Given a new house features, predicts its approximate selling price

Output is continuous value (e.g., ₹25,50,000)

Comparison Table:

Aspect Classification Regression

Output Type Categorical Continuous

Output Possible Values Fixed categories Any numerical value in range

Example Output "Spam" or "Not Spam" 25,50,000 rupees

Model Type Discrete Continuous

Use Case Email filtering, Disease diagnosis Price prediction, Weather forecasting
Q4. (5 Marks) What is Clustering in Unsupervised Learning? How is it Different from
Classification? Explain with a Real-World Example
Marking Scheme:

Definition of Clustering (1 mark)

Definition of Classification (0.5 mark)

Key differences (1.5 marks)

Real-world example (2 marks)

Answer:

Clustering:

Definition: Unsupervised learning technique that groups similar data points together. No
predefined classes; machine discovers groupings based on data similarity.

Characteristics:
Works with unlabeled data
No predefined categories

Groups data based on similarities

Number of clusters can vary

Classification:

Definition: Supervised learning technique that assigns data to predefined classes. Uses labeled
training data.

Key Differences:

Aspect Clustering Classification

Type of Learning Unsupervised Supervised

Data Requirement Unlabeled Labeled

Classes No predefined classes Predefined classes

Process Discovers groupings Assigns to known classes

Method Similarity-based Rule/pattern-based from training

Real-World Example – Customer Segmentation:

Scenario: E-commerce company has customer purchase history but no predefined segments.

Clustering Approach:

Collect customer data (purchase history, frequency, amount spent)

No labels telling "high-value" or "low-value" customer

Clustering algorithm analyzes patterns

Discovers customers naturally group into:

 Cluster 1: Frequent buyers, high spending

Cluster 2: Occasional buyers, moderate spending

Cluster 3: Rare buyers, low spending
Company creates targeted marketing for each cluster
If It Were Classification:

Company would need historical data pre-labeled as "high-value," "medium-value," "low-

value"

Train model on these labeled examples

Apply model to new customers

Assigns new customers to known categories

Q5. (5 Marks) Explain Neural Network Structure and How AI Makes Decisions Using
Perceptron Concept. Illustrate with an Example
Marking Scheme:

Definition of Neural Network (1 mark)

Explanation of three layers (1.5 marks)

Perceptron concept explanation (1 mark)
Example with calculation (1.5 marks)

Answer:

Neural Network Structure:

Definition: System of organizing machine learning algorithms modeled after brain neurons.
Automatically extracts features from data.

Three Main Layers:

1. Input Layer:

First layer that receives raw data

Acquires information and feeds to network

No processing occurs here

Acts as data entry point

2. Hidden Layer(s):
Performs all computations and processing

Each node processes data using weights and biases

Contains machine learning algorithms

Multiple hidden layers can exist for complex functions

Extracts features progressively

 3. Output Layer:

Receives processed information from hidden layers

Produces final decision/classification
No processing occurs
User interface for results

Perceptron – How AI Makes Decisions:

Definition: Basic unit mimicking brain neuron; takes weighted inputs and produces output.

Process:
1. Collect inputs (X1, X2, X3, ...)

2. Assign weights (W1, W2, W3, ...) representing importance

3. Calculate: Sum = (X1×W1) + (X2×W2) + ... + (Bias×WBias)

4. Compare with threshold
5. Output decision: If Sum > Threshold, then YES; else NO

Example – Deciding to Play Cricket:

Inputs (Factors):

X1: Is weather sunny? (1=yes, 0=no)

X2: Do I have free time? (1=yes, 0=no)
X3: Are friends available? (1=yes, 0=no)

Weights (Importance):

W1 = 2 (weather most important)

W2 = 1.5 (free time)

W3 = 1 (friends availability)
Bias = -2 (tendency not to play)

Scenario: Weather is sunny (1), I have free time (1), friends available (1)

Calculation:

Sum = (1×2) + (1×1.5) + (1×1) + (-2) = 2 + 1.5 + 1 - 2 = 2.5

Threshold = 0
Since 2.5 > 0 → Decision: YES, PLAY CRICKET

Another Scenario: Weather not sunny (0), no free time (0), friends available (1)

Calculation:

Sum = (0×2) + (0×1.5) + (1×1) + (-2) = 0 + 0 + 1 - 2 = -1

Threshold = 0

Since -1 < 0 → Decision: NO, DON'T PLAY CRICKET

Key Points:

Different people assign different weights based on preferences

Bias adjusts baseline tendency
No "right" weights – depends on individual values

This is how neural networks learn: adjusting weights to match desired outputs