ML_Insem

UNIT 1
## Explain Machine Learning Models (Easy & Understandable Language)

What is a Machine Learning Model?

A Machine Learning Model is a method or technique that helps a computer learn patterns from data and make
predictions or decisions without being explicitly programmed.

Example:
If you give a model thousands of emails labeled as Spam and Not Spam, it learns the pattern and can identify new
spam emails.

There are six common Machine Learning models:

1. Geometric Model

A Geometric Model represents data as points in space (2D, 3D, or multi-dimensional). It uses the distance between
data points to classify or group them.

Simple Idea

👉 Similar data points stay close together.

👉 Different data points stay far apart.

Example

Suppose we have students' data:

Study Hours Result

2 Fail

3 Fail

8 Pass

9 Pass

If we plot these points on a graph:

 Students with similar study hours are close.

 A new student who studies 8.5 hours is near the "Pass" group, so the model predicts Pass.

Advantages

 Easy to understand.

 Good for classification problems.

Disadvantages

 Not suitable for very complex data.

2. Probabilistic Model

A Probabilistic Model predicts outcomes using probability.

It answers:

"What is the chance that this belongs to a particular class?"

ML_Insem
Simple Idea

Instead of saying "This email is spam,"

it says

Spam = 95%
Not Spam = 5%

Example

Suppose an email contains the words:

 Win Money

 Free Gift

The model calculates

Probability(Spam) = 96%

So it classifies the email as Spam.

Advantages

 Gives confidence level.

 Handles uncertainty well.

Disadvantages

 Requires probability calculations.

3. Statistical Model

A Statistical Model uses mathematical and statistical methods to find relationships between data.

Simple Idea

It studies data and finds patterns using statistics.

Example

Suppose we have

Study Hours Marks

2 30

4 50

6 70

8 90

The model finds

More study hours → Higher marks

It can predict marks for a student who studies 5 hours.

Examples of Statistical Models

 Linear Regression
ML_Insem
 Logistic Regression

Advantages

 Good prediction accuracy.

 Easy to interpret.

Disadvantages

 Assumes relationships follow statistical rules.

4. Logical Model

A Logical Model makes decisions using IF–THEN rules.

Simple Idea

It works like human reasoning.

Example

IF Temperature > 38°C

THEN Patient has Fever

IF Fever AND Cough

THEN Flu

Another Example

IF Age > 18

THEN Eligible to Vote

ELSE Not Eligible

Decision Trees are common logical models.

Advantages

 Very easy to understand.

 Easy to explain decisions.

Disadvantages

 Large rule sets become complicated.

5. Grouping Model

Definition

A Grouping Model divides similar data into groups (clusters).

It does not require labeled data.

Simple Idea
ML_Insem
Similar objects are placed in the same group.

Example

A shopping website groups customers into:

Group 1 → Students

Group 2 → Office Employees

Group 3 → Senior Citizens

Each group has similar buying habits.

Common Algorithm

 K-Means Clustering

Advantages

 Finds hidden patterns.

 Useful for customer segmentation.

Disadvantages

 Number of groups often needs to be chosen in advance.

6. Grading Model

Definition

A Grading Model assigns a score, rank, or grade instead of simply classifying data.

Simple Idea

Instead of saying Good or Bad, it gives a rating.

Example

Movie Recommendation

Movie Rating

Movie A ⭐⭐⭐⭐⭐

Movie B ⭐⭐⭐⭐

Movie C ⭐⭐⭐

Or in education:

Marks Grade

90 A

75 B

60 C

Advantages

 Useful for ranking.

ML_Insem
 Helps compare items.

Disadvantages

 Choosing the grading scale can be difficult.

Artificial Intelligence (AI) Machine Learning (ML)

AI is the science of making machines intelligent. ML is a branch of AI that helps machines learn from data.

AI aims to mimic human intelligence. ML aims to identify patterns and make predictions.

AI can work using rules, logic, and learning

ML mainly works by learning from data using algorithms.
techniques.

AI does not always require large amounts of

ML requires training data to learn.
data.

AI solves a wide variety of intelligent tasks. ML mainly focuses on prediction and decision-making.

AI is a broader concept. ML is a subset of AI.

Examples: Robots, Virtual Assistants, Self-driving Examples: Spam detection, Recommendation systems, Image
cars. recognition.

## Types of Machine Learning

1. Supervised Learning

In Supervised Learning, the computer learns using labeled data.

Labeled data means the correct answer is already given.

👉 It is like a teacher teaching a student.

Example

A teacher gives students questions with answers.

Similarly,

Email Label

"Win ₹1 Crore" Spam

"Meeting at 10 AM" Not Spam

The computer learns from these examples.

Now, if a new email says:

"Congratulations! You won ₹5 lakh."

The computer predicts:

Spam

Real-Life Examples
ML_Insem
 Spam Email Detection

 House Price Prediction

 Student Result Prediction

 Disease Prediction

Advantages

 High accuracy

 Easy to train

 Gives good predictions

Disadvantages

 Needs a large amount of labeled data.

 Preparing labeled data takes time.

2. Unsupervised Learning

In Unsupervised Learning, the computer learns using unlabeled data.

There are no correct answers.

The computer finds similar data and makes groups by itself.

👉 It is like asking students to form groups on their own without any instructions.

Example

A shopping mall has customer information.

The computer automatically groups customers into:

 Students

 Office Employees

 Senior Citizens

No one tells the computer which customer belongs to which group.

Real-Life Examples

 Customer Segmentation

 Product Recommendation

 Finding Similar Images

Advantages

 No labeled data required.

 Finds hidden patterns.

Disadvantages

 Results may not always be accurate.

ML_Insem
 Difficult to understand the groups.

3. Semi-supervised Learning

Semi-supervised Learning uses both labeled and unlabeled data.

Only a small amount of data has labels, while the remaining data has no labels.

👉 It is a combination of Supervised and Unsupervised Learning.

Example

Suppose you have 1000 photos.

 100 photos are labeled as Cat or Dog.

 900 photos have no labels.

The computer learns from the 100 labeled photos and then predicts labels for the remaining 900 photos.

Real-Life Examples

 Face Recognition

 Medical Image Analysis

 Speech Recognition

Advantages

 Needs fewer labeled data.

 More accurate than unsupervised learning.

 Saves time and cost.

Disadvantages

 More difficult to implement.

 Accuracy depends on the labeled data.

4. Reinforcement Learning

Definition

In Reinforcement Learning, the computer learns by trying different actions.

If it makes the correct decision, it gets a Reward.

If it makes the wrong decision, it gets a Penalty.

👉 It learns by trial and error, just like humans.

Example

A robot is learning to walk.

 Walks correctly → Reward ✅

 Falls down → Penalty ❌

After many attempts, the robot learns the correct way to walk.
ML_Insem
Real-Life Examples

 Self-Driving Cars

 Chess Playing AI

 Robots

 Video Games

Advantages

 Learns from experience.

 Improves continuously.

 Good for decision-making tasks.

Disadvantages

 Takes a long time to train.

 Needs many trials.

 Can be expensive.

Q. What is Statistical Learning?

Statistical Learning is a method in Machine Learning that uses statistics (mathematics) to understand data, find
patterns, and make predictions.

Statistical Learning is the process of learning from data using statistical methods to predict future results or
make decisions.

Real-Life Example

Suppose a teacher wants to predict a student's marks.

He collects:

Study Hours Marks

2 35

4 55

6 75

8 92

After observing this data, he finds a pattern:

👉 More study hours → Higher marks

Now, if a student studies 5 hours, the model can predict that the student may score around 65–70 marks.

This is Statistical Learning.

Types of Statistical Learning Approaches

There are two main approaches:

1. Parametric Approach
ML_Insem
A Parametric Approach assumes that the relationship between input and output follows a fixed mathematical
formula.

The model has a fixed number of parameters, so it is simple and fast.

Simple Example

Suppose we want to predict marks based on study hours.

The model assumes:

Marks = a × Study Hours + b

where a and b are fixed values learned from the data.

Once these values are found, the model can predict marks for new students.

Real-Life Example

Predicting:

 House Prices

 Student Marks

 Salary

Advantages

 Easy to understand.

 Fast training.

 Works well with small datasets.

Disadvantages

 Assumes a fixed relationship.

 May not work well for complex data.

Examples

 Linear Regression

 Logistic Regression

2. Non-Parametric Approach

A Non-Parametric Approach does not assume any fixed mathematical formula.

It learns patterns directly from the data, making it more flexible.

Simple Example

Suppose you want to identify different types of fruits.

Instead of using a fixed formula, the model looks at:

 Color

 Shape

 Size
ML_Insem
 Weight

It learns from these features and classifies the fruits.

Real-Life Example

 Face Recognition

 Image Classification

 Speech Recognition

 Recommendation Systems

Advantages

 Can handle complex data.

 More flexible.

 Often gives better accuracy with large datasets.

Disadvantages

 Requires more data.

 Takes longer to train.

 More computationally expensive.

Examples

 Decision Trees

 K-Nearest Neighbors (KNN)

 Random Forest

Q. What is a Data Format?

A Data Format is the way data is organized and stored so that a Machine Learning model can understand and
process it.

Simple Definition

Data Format is the structure in which data is stored before it is given to a Machine Learning model.

Example

Student data:

Name Age Marks

Rahul 18 85

Priya 19 90

This is one type of data format (Table format).

Types of Data Formats in Machine Learning

1. Structured Data
ML_Insem
Structured Data is data that is organized in rows and columns.

It is easy to store, search, and process.

Example

Student Age Marks

Rahul 18 85

Priya 19 90

This type of data is stored in:

 Excel

 Database

 CSV files

Applications

 Student Records

 Bank Records

 Employee Database

Advantages

 Easy to understand.

 Easy to analyze.

 Fast processing.

2. Semi-Structured Data

Semi-Structured Data does not follow a strict table format but has tags or labels that organize the data.

Example (JSON)

"Name": "Rahul",

"Age": 18,

"Marks": 85

or XML:

<Student>

<Name>Rahul</Name>

<Age>18</Age>

</Student>

Applications
ML_Insem
 APIs

 Web Applications

 Mobile Apps

Advantages

 Flexible.

 Easy to exchange data between applications.

3. Unstructured Data

Unstructured Data has no fixed format.

It cannot be stored easily in rows and columns.

Examples

 Images

 Videos

 Audio

 Emails

 PDF Documents

 Social Media Posts

Applications

 Face Recognition

 Speech Recognition

 Chatbots

 Video Analysis

Advantages

 Contains rich information.

 Used in AI applications.

Disadvantages

 Difficult to analyze.

 Requires preprocessing before using in ML.

4. Multidimensional Data

Multidimensional Data contains many features (attributes) for each data item.

Each feature represents one dimension.

Example

Suppose a student record contains:

ML_Insem
 Name

 Age

 Height

 Weight

 Attendance

 Marks

 Sports Score

Each feature is a dimension.

So the student data becomes multidimensional.

Applications

 Medical Diagnosis

 Image Processing

 Data Mining

 Machine Learning

Advantages

 Stores more information.

 Improves prediction accuracy.

Disadvantages

 More difficult to process.

 Needs feature selection or dimensionality reduction (such as PCA).

ML_Insem

UNIT 2
Q. Principal Component Analysis (PCA)

Principal Component Analysis (PCA) is a feature extraction and dimensionality reduction technique used in
Machine Learning.

It reduces the number of features (variables) in a dataset while keeping most of the important information.

Simple Definition

PCA is a technique that converts many related features into a smaller number of new features (called
Principal Components) without losing much important information.

Why Do We Need PCA?

Sometimes a dataset has too many features, making it:

 Slow to process.

 Difficult to analyze.

 More likely to contain duplicate or unnecessary information.

PCA solves this problem by reducing the number of features while preserving most of the useful data.

Easy Example

Suppose a student database contains:

 Name
 Roll No.
 Age
 Height
 Weight
ML_Insem
 Attendance
 Internal Marks
 Semester Marks
 Final Marks

There are many features.

PCA combines the related features into a few important Principal Components, making the data simpler and faster
to process.

Think of it like summarizing a long chapter into short notes—you keep the important points and remove
unnecessary details.

PCA Algorithm (Easy Steps)

Step 1: Collect the Data

Gather the dataset that contains many features.

Step 2: Standardize the Data

Convert all features to the same scale so that no feature dominates because of its larger values.

Step 3: Calculate the Covariance Matrix

Find how different features are related to each other.

Step 4: Find Eigenvalues and Eigenvectors

 Eigenvalues tell us how much information a component contains.

 Eigenvectors show the direction of the new principal components.

Step 5: Select Principal Components

Choose the components with the highest Eigenvalues, as they contain the most useful information.

Step 6: Transform the Data

Convert the original data into the selected principal components.

The dataset now has fewer features but still retains most of the important information.

Applications of PCA

1. Image Compression

Reduces image size while maintaining image quality.

2. Face Recognition

Removes unnecessary image information and focuses on important facial features.

3. Medical Data Analysis

Reduces the number of medical attributes for faster diagnosis.

4. Data Visualization

Converts high-dimensional data into 2D or 3D for easier understanding.

5. Noise Removal

Eliminates unwanted or irrelevant information from data.

ML_Insem
## Feature Selection

Q. What is Feature Selection?

Feature Selection is the process of selecting only the important features (columns) from a dataset and removing
unnecessary or irrelevant features.

Feature Selection is the process of choosing the most useful features from a dataset to improve the
performance of a Machine Learning model.

Why is Feature Selection Needed?

Sometimes a dataset contains unnecessary columns.

For example:

Name Roll No Age Height Weight Marks

If we want to predict Marks, then:

 ✔ Age may be useful.

 ✔ Study Hours may be useful.

 ✖ Name is usually not useful.

 ✖ Roll Number is usually not useful.

Feature Selection removes unnecessary columns and keeps only useful ones.

Feature Selection Methods

There are three main methods:

1. Filter Method

2. Wrapper Method

3. Embedded Method

1. Filter Method

The Filter Method selects features before training the Machine Learning model.

It uses statistical techniques to find the most important features.

The model is not involved in selecting the features.

Simple Idea

👉 First, remove unnecessary features.

👉 Then train the model.

Example

Suppose we have:

Feature Importance

Study Hours High

ML_Insem
Feature Importance

Attendance High

Name Very Low

Roll Number Very Low

The Filter Method removes Name and Roll Number before training the model.

2. Wrapper Method

The Wrapper Method selects features by training the Machine Learning model multiple times with different
feature combinations.

It chooses the combination that gives the best accuracy.

Simple Idea

👉 Try different feature combinations.

👉 Keep the combination with the best result.

Example

Suppose we have four features:

 Age

 Height

 Weight

 Attendance

The Wrapper Method tries:

 Age + Attendance

 Height + Weight

 Age + Height + Attendance

It compares the accuracy of each combination and selects the best one.

3. Embedded Method

The Embedded Method performs feature selection while training the Machine Learning model.

Unlike the Filter Method (before training) and Wrapper Method (repeated training), feature selection happens during
model training.

Example

Some algorithms automatically identify important features while learning.

Examples:

 Decision Tree

 Random Forest
ML_Insem
 LASSO Regression

Q. What is Data Scaling?

Data Scaling is the process of changing the values of data into a common range without changing their meaning.

Data Scaling is a technique used to make all features have a similar scale so that no feature has more
importance just because it has larger values.

Normalization

Normalization is a type of Data Scaling.

It converts data into a small and common range, usually 0 to 1.

The two most common methods are:

1. Min-Max Scaling

2. Z-score Normalization (Standardization)

1. Min-Max Scaling

Min-Max Scaling converts the data into a range of 0 to 1.

The smallest value becomes 0, and the largest value becomes 1.

Formula

Value−Minimum
New Value=
Maximum−Minimum
Example

Given values:

10, 20, 30, 40, 50

Minimum = 10

Maximum = 50

Find the normalized value of 30.

Solution

30−10 20
¿ ¿ ¿ 0.5
50−10 40
Answer: 0.5

2. Z-score Normalization (Standardization)

Z-score Normalization converts data so that it has:

 Mean = 0

 Standard Deviation = 1

It tells us how far a value is from the average.

ML_Insem
Formula

X−μ
Z=
σ
Where:

 X = Data value

 μ (Mean) = Average of all values

 σ (Standard Deviation) = Spread of the data

Example

Given:

Value (X) = 70

Mean (μ) = 60

Standard Deviation (σ) = 5

Solution

70−60 10
Z= Z= Z=2
5 5
Answer: 2

This means the value 70 is two standard deviations above the average.

Q. Local Binary Pattern (LBP)

What is Local Binary Pattern (LBP)?

Local Binary Pattern (LBP) is a feature extraction technique used in image processing and machine learning.

It is used to identify the texture (surface pattern) of an image by comparing each pixel with its neighboring pixels.

Simple Definition

Local Binary Pattern (LBP) is a method used to describe the texture of an image by comparing the center
pixel with its surrounding pixels.

Why is LBP Used?

LBP helps computers recognize patterns in images.

It is mainly used for:

 Face Recognition

 Fingerprint Recognition

 Image Classification

 Medical Image Analysis

 Texture Detection
ML_Insem
How Does LBP Work?

LBP compares the center pixel with its 8 neighboring pixels.

Rule

 If the neighboring pixel is greater than or equal to the center pixel → Write 1

 If the neighboring pixel is less than the center pixel → Write 0

These 8 binary digits form a binary number.

Finally, the binary number is converted into a decimal number, called the LBP value.

Steps of LBP Algorithm

Step 1

Select the center pixel.

Step 2

Compare each surrounding pixel with the center pixel.

Step 3

Write:

 1 if Neighbor ≥ Center

 0 if Neighbor < Center

Step 4

Arrange the binary values in a fixed order (usually clockwise).

Step 5

Convert the binary number into a decimal value.

This decimal value is the LBP code.

Numerical Example

Consider the following 3 × 3 pixel matrix.

20 30 10

40 25 35

15 50 28

Step 1

Center Pixel = 25

Step 2

Compare every neighboring pixel with 25.

ML_Insem
Neighbor Compare with 25 Binary

20 20 < 25 0

30 30 ≥ 25 1

10 10 < 25 0

35 35 ≥ 25 1

28 28 ≥ 25 1

50 50 ≥ 25 1

15 15 < 25 0

40 40 ≥ 25 1

Step 3

Binary Pattern (Clockwise)

010

1 1

101

Binary Number:

01011101

Step 4

Convert Binary to Decimal

01011101₂

= 93

✅ LBP Value = 93

Applications of LBP

1. Face Recognition

Used to identify people based on facial texture.

2. Fingerprint Recognition

Detects fingerprint patterns.

3. Medical Imaging

Helps detect diseases from X-ray or MRI images.

4. Image Classification

Groups images into different categories.

5. Texture Analysis

Identifies different surface textures such as wood, cloth, or stone.

ML_Insem