1. What is human learning? Give any two examples.

Ans Human learning is a complex and ongoing process through which individuals acquire knowledge,
skills, attitudes, or behaviors. It occurs as a result of experience, observation, practice, study, or
teaching. Learning is not just about memorizing facts—it involves understanding, applying, and
sometimes even creating new ideas. It leads to a relatively permanent change in behavior or
thinking, meaning that what is learned can be applied in real-life situations and retained over time.

Human learning can be formal, such as in schools and colleges, where structured lessons and
assessments are used, or informal, like learning through everyday life experiences, observation, or
problem-solving. It involves cognitive, emotional, and social aspects, and each individual may learn
differently depending on their abilities, motivation, and environment.

Examples of Human Learning:

-Learning to ride a bicycle: This is a skill-based learning where a person gains balance, coordination,
and control through repeated practice and experience. Once learned, the person can ride a bicycle
for years without forgetting, demonstrating permanent change.

-Learning a new language: This involves acquiring knowledge of vocabulary, grammar, pronunciation,
and cultural nuances through listening, speaking, reading, and writing. Over time, a person becomes
capable of communicating fluently in that language, showing both cognitive and behavioral learning.

2. What is machine learning? What are key tasks of machine learning?

Ans Machine Learning (ML)

Machine Learning is a branch of artificial intelligence (AI) that enables computers and systems to
learn from data, identify patterns, and make decisions or predictions without being explicitly
programmed. Instead of following strict rules coded by a human, a machine learning model improves
its performance over time as it is exposed to more data.

In simpler terms, machine learning allows computers to “learn from experience” much like humans
do, but using mathematical algorithms and statistical techniques.

Key Tasks of Machine Learning

Machine learning tasks are generally divided into the following main types:

-Supervised Learning

The model is trained on labeled data, meaning the input comes with the correct output.

Goal: Predict outcomes for new, unseen data.

Examples:

Predicting house prices based on features like size and location.

Spam email detection (emails labeled as “spam” or “not spam”).

-Unsupervised Learning

The model is trained on unlabeled data and must find patterns, relationships, or structures on its
own.
Goal: Group or organize data meaningfully.

Examples:

Customer segmentation in marketing.

Grouping similar news articles together.

-Reinforcement Learning

The model learns by trial and error, receiving rewards or penalties based on its actions.

Goal: Maximize cumulative reward over time.

Examples:

Training a robot to walk or perform tasks.

Game-playing AI like AlphaGo.

-Semi-supervised Learning (Combination of supervised & unsupervised)

Uses a small amount of labeled data and a large amount of unlabeled data to improve learning
accuracy.

Example: Facial recognition systems trained with a few labeled images but thousands of unlabeled
ones.

✅ In short: Machine learning is about teaching machines to learn patterns from data, and its key tasks
involve prediction, classification, clustering, and decision-making.

3. What are different objectives of machine learning? How are these

related with human learning?

Ans.### **Objectives of Machine Learning**

The main objectives of **machine learning (ML)** are:

1. **Prediction**

* ML models aim to predict outcomes or future trends based on past data.

* **Example:** Predicting stock prices, weather forecasts, or patient disease risk.

2. **Classification**

* ML models categorize data into predefined classes or groups.

* **Example:** Email spam detection (spam vs. not spam), tumor type classification (benign vs.
malignant).

3. **Pattern Recognition**

* ML identifies patterns and regularities in data that may not be obvious to humans.

* **Example:** Handwriting recognition, speech recognition, customer behavior patterns.

4. **Clustering / Grouping**

* ML groups similar data points together without prior labels.

* **Example:** Market segmentation, grouping similar images or products.

5. **Decision Making / Optimization**

* ML helps make decisions by analyzing data and choosing the best possible action.

* **Example:** Self-driving cars choosing optimal routes, recommendation systems suggesting

products.

6. **Automation / Self-Improvement**

* ML systems can **improve automatically** from experience without explicit programming.

* **Example:** Google’s search algorithm improves over time by learning user behavior.

### **Relation to Human Learning**

Machine learning objectives closely **mirror human learning processes**:

| **Human Learning** | **Machine Learning** |

| ---------------------------------------------- | -------------------------------------------------- |

| Learn from experience and past outcomes | Predict outcomes from historical data |

| Recognize patterns (faces, language, behavior) | Pattern recognition in data |

| Categorize objects or information | Classification of data into groups |

| Solve problems and make decisions | Optimization and decision-making algorithms |

| Learn through trial and error | Reinforcement learning models improve via feedback |

| Improve skills over time | ML models improve automatically with more data |

✅ **In short:** Machine learning is like teaching a computer to **think, learn, and improve** in
ways similar to humans, but using **mathematical algorithms and data** instead of the human
brain.

4. Define machine learning and explain the different elements with a real example.

Ans Definition:

Machine Learning (ML) is a branch of Artificial Intelligence (AI) that allows computers to learn from
data and experience without explicit programming. It uses algorithms to identify patterns, make
predictions, or take actions based on data.

Elements of Machine Learning:

Data – The raw information used to train the ML model.

Example: In a house price prediction model, data includes size, location, number of rooms, age of the
house, etc.

Model – A mathematical representation of a problem that makes predictions or decisions.

Example: A linear regression model predicts house prices based on input features.

Features – Individual measurable properties or characteristics used as input.

Example: In house price prediction, features are size ([Link].), number of bedrooms, and location.

Learning Algorithm – The method used to train the model and find patterns in the data.

Example: Linear regression algorithm, decision trees, or neural networks.

Target / Label – The outcome or value the model is trying to predict.

Example: The actual house price.

Prediction / Output – The result provided by the trained model for new input data.

Example: For a 1200 [Link]. house in a city, the model predicts the price as $150,000.

Real Example: Predicting house prices:

Input: Size, location, number of rooms

Output: Predicted price

Algorithm: Linear regression

Model learns from historical house data to make accurate prediction.

5. Explain the process of abstraction with an example.

Ans Process of Abstraction

Abstraction is the process of simplifying a complex system or problem by focusing only on the
essential details while ignoring irrelevant or less important information. In other words, it helps us
reduce complexity and deal with only the important aspects of a problem.

In computer science and machine learning, abstraction allows us to create models, algorithms, or
systems without getting overwhelmed by unnecessary details.

Steps in the Process of Abstraction

-Identify the problem – Understand what needs to be solved.

-Determine essential features – Decide which aspects are important for solving the problem.

-Ignore irrelevant details – Remove information that doesn’t affect the outcome.

-Create a simplified model – Use only the relevant features to represent the problem.

-Solve or analyze the problem – Work with the simplified model to find a solution.

Example of Abstraction

Scenario: Designing a self-driving car.

Complex Reality: A real environment has cars, pedestrians, traffic signals, road signs, weather
conditions, road bumps, birds, buildings, etc.

Abstraction: Focus only on the essential information needed for safe driving:
Road lanes (to stay on track)

Traffic lights (to stop or go)

Other vehicles (to avoid collisions)

Pedestrians (to prevent accidents)

The self-driving car ignores irrelevant details like the color of buildings, trees, or advertisements,
making the problem simpler and solvable.

Result: The car can make decisions efficiently using only the important information.

6. What is generalization? What role does it play in the process of machine learning?

Ans Generalization in machine learning refers to the ability of a model to perform well on new,
unseen data, not just the data it was trained on. When a machine learning model learns from a
dataset, it tries to identify patterns that represent the underlying relationship in the data. If the
model memorizes every detail of the training data, it may perform perfectly on that data but fail to
make accurate predictions on new data; this problem is called overfitting. On the other hand, if the
model fails to capture the important patterns, it will perform poorly even on the training data, which
is called underfitting. Generalization is achieved when the model captures the essential patterns
without memorizing noise, allowing it to make correct predictions on data it has never seen before.

For example, consider a model trained to recognize cats in images. If it has good generalization, it can
correctly identify cats in new images that were not part of its training dataset. If the model only
memorized the specific images it was trained on, it might fail when it encounters cats in different
poses, backgrounds, or lighting conditions. The ability to generalize is what makes machine learning
models useful in real-world applications, such as predicting house prices, detecting spam emails, or
recognizing speech. Without generalization, a model’s predictions would be unreliable outside the
training environment, limiting its practical value.

7. Explain the concept of penalty and reward in reinforcement learning.

Ans In reinforcement learning (RL), an agent learns to make decisions by interacting with its
environment. The agent performs actions, observes the results, and receives feedback that helps it
learn which actions are good or bad. This feedback comes in the form of rewards and penalties.

A reward is a positive signal given to the agent when it takes a correct or desirable action. It
encourages the agent to repeat that action in the future. A penalty (or negative reward) is given
when the agent takes a wrong or undesirable action. It discourages the agent from repeating that
action. Over time, by trying actions, receiving rewards, and avoiding penalties, the agent learns the
best strategy to maximize its total reward.

Example: Imagine a robot learning to navigate a maze. If the robot moves closer to the exit, it
receives a reward (+10 points). If it bumps into a wall, it gets a penalty (-5 points). By repeating this
process, the robot learns the most efficient path to reach the exit while avoiding walls.

In short, rewards guide the agent toward correct actions, and penalties help it avoid mistakes,
allowing it to learn optimal behavior through trial and error.

8. What is regression? Give example of some practical problems solved using regression.
Ans Regression is a type of supervised machine learning used to predict a continuous numerical
value based on input data. Unlike classification, which predicts categories, regression predicts
quantities that can take any value within a range. The model learns patterns from historical data and
uses them to estimate future or unknown values.

Practical examples of problems solved using regression:

Predicting house prices based on features like size, location, number of bedrooms, and age of the
house.

Forecasting stock market prices using historical price trends and economic indicators.

Predicting temperature or rainfall based on past weather data.

Estimating sales revenue based on advertising expenditure or seasonal trends.

Predicting a person’s weight based on height, age, and other health metrics.

In all these examples, regression helps quantify relationships between variables and make
predictions that guide decisions in real-world scenarios.

9. What is classification? Explain the key differences between classification and regression.

Ans Classification is a type of supervised machine learning where the model learns from labeled data
to assign new inputs into predefined categories or classes. The output is always discrete, meaning it
belongs to one of the set of classes.

Example:

Detecting whether an email is “Spam” or “Not Spam.”

Predicting whether a tumor is “Benign” or “Malignant.”

| **Feature** | **Classification** | **Regression** |

| ---------------------- | ----------------------------------- | ----------------------------------------------- |

| **Output Type** | Discrete / categories | Continuous / numerical |

| **Goal** | Assign input to a class | Predict a numeric value |

| **Examples** | Spam detection, disease diagnosis | House price prediction, temperature

forecasting |

| **Evaluation Metrics** | Accuracy, Precision, Recall | Mean Squared Error (MSE), R², MAE
|

| **Question Answered** | “Which category does it belong to?” | “What is the value?”
|

10. Explain the process of clustering in details.

Ans Clustering is an unsupervised machine learning technique that groups similar data points
together into clusters based on their characteristics, without using any labeled data. The objective is
to ensure that data points within the same cluster are more similar to each other than to those in
different clusters, enabling the discovery of natural groupings and hidden patterns in complex
datasets.

Goal: Discover the natural grouping or structure in unlabeled data without predefined categories.

How: Data points are assigned to clusters based on similarity or distance measures.

Similarity Measures: Can include Euclidean distance, cosine similarity or other metrics depending on
data type and clustering method.

Output: Each group is assigned a cluster ID, representing shared characteristics within the cluster.

11. Write a short note on Application of machine learning algorithms.

Ans Machine Learning (ML) algorithms are widely used to enable systems to learn from data and
make intelligent decisions without explicit programming. Some key applications include:

Healthcare: Disease prediction, medical image analysis, personalized treatment recommendations.

Finance: Fraud detection, credit scoring, stock price prediction, risk assessment.

Marketing: Customer segmentation, recommendation systems, sentiment analysis, targeted

advertising.

Education: Student performance prediction, adaptive learning systems.

Transportation: Traffic prediction, self-driving cars, route optimization.

E-commerce: Product recommendations, demand forecasting, price optimization.

Security: Face recognition, spam filtering, intrusion detection systems.

Overall, machine learning algorithms improve efficiency, accuracy, and automation across various
industries.

12. Write a short note on Supervised learning.

Ans Supervised learning is a type of machine learning in which a model is trained using labeled data.
Each training example consists of an input and a corresponding correct output (label). The algorithm
learns a mapping between inputs and outputs and uses this learned relationship to make predictions
on new, unseen data.

Common supervised learning tasks include classification (e.g., spam email detection, disease
diagnosis) and regression (e.g., house price prediction, sales forecasting). Popular supervised
learning algorithms include Linear Regression, Logistic Regression, Decision Trees, Support Vector
Machines (SVM), and k-Nearest Neighbors (k-NN).

Supervised learning is widely used because of its high accuracy when sufficient labeled data is
available.

[Link] a short note on Unsupervised learning.

Ans Unsupervised learning is a type of machine learning where the model is trained on unlabeled
data, meaning no predefined output or target is provided. The algorithm discovers hidden patterns,
structures, or relationships within the data on its own.
Common unsupervised learning tasks include clustering (e.g., customer segmentation, grouping
similar data points) and association (e.g., market basket analysis). Popular unsupervised learning
algorithms include K-Means clustering, Hierarchical clustering, DBSCAN, and Apriori algorithm.

Unsupervised learning is widely used in data exploration, pattern recognition, anomaly detection,
and feature extraction, helping organizations gain insights from large datasets.

14. Write a short note on Reinforcement learning.

[Link] learning is a type of machine learning in which an agent learns by interacting with
an environment. The agent takes actions and receives rewards or penalties as feedback, with the
goal of maximizing cumulative reward over time.

Unlike supervised learning, reinforcement learning does not use labeled data. Instead, learning is
based on trial and error. Common components include the agent, environment, actions, states, and
reward. Popular reinforcement learning algorithms include Q-learning, SARSA, and Deep Q-Networks
(DQN).

Reinforcement learning is widely used in robotics, game playing (e.g., AlphaGo), autonomous
vehicles, and recommendation systems, where sequential decision-making is required.

15. Write a difference between abstraction and generalization.

Ans Below is a detailed difference between Abstraction and Generalization, explained clearly for
exams and understanding:

Detailed Difference Between Abstraction and Generalization

Aspect Abstraction Generalization

Process of hiding internal Process of identifying common

Definition implementation details and showing only characteristics among classes and
essential features creating a general class

Focuses on what objects have in

Focus Focuses on what an object does
common

Improve reusability and reduce

Objective Reduce complexity and increase clarity
redundancy

Level Conceptual / design-level concept Structural / relationship-based concept

Achieved using abstract classes and

Implementation Achieved using inheritance
interfaces

Moves from specific classes to a general

Direction Moves from details to essential behavior
class

Data Visibility Internal data is hidden Common data and methods are shared

Flexibility Provides flexibility in implementation Provides code reuse and hierarchy

Example A Vehicle interface defines start() and Car and Bike are generalized into a
Aspect Abstraction Generalization

stop() without implementation Vehicle class

Object-oriented design for simplicity and Object-oriented design for hierarchy and
Used In
security reuse

In Short

 Abstraction = Hiding details, showing essentials

 Generalization = Combining similar entities into a common parent

16. Write a difference between supervised and unsupervised learning.

17. Write a difference between classification and regression.

Ans Difference Between Classification and Regression

Aspect Classification Regression

Type of Output Discrete / Categorical values Continuous / Numerical values

Aspect Classification Regression

Purpose Assigns input data to predefined classes Predicts a numerical value

Examples of Spam / Not Spam, Yes / No, Disease / No

Price, Temperature, Sales amount
Output Disease

Common Logistic Regression, Decision Tree, Naïve Linear Regression, Polynomial Regression,
Algorithms Bayes, SVM, k-NN SVR, Decision Tree

Evaluation
Accuracy, Precision, Recall, F1-score Mean Squared Error (MSE), RMSE, R²
Metrics

Email spam detection, image

Use Case House price prediction, rainfall prediction
classification

In short:

 Classification predicts categories

 Regression predicts numbers

18. What are the main activities involved in machine learning? What is meant by data pre-
processing?

Ans Main Activities Involved in Machine Learning

The main activities involved in machine learning are:

1. Data Collection – Gathering relevant data from various sources.

2. Data Pre-processing – Cleaning and preparing data for use.

3. Feature Selection / Feature Engineering – Selecting important features or creating new

ones.

4. Model Selection – Choosing a suitable machine learning algorithm.

5. Model Training – Training the model using prepared data.

6. Model Evaluation – Measuring performance using evaluation metrics.

7. Deployment / Prediction – Applying the trained model to real-world data.

Data Pre-processing

Data pre-processing is the process of cleaning, transforming, and organizing raw data into a suitable
format for machine learning algorithms. Real-world data often contains missing values, noise, and
inconsistencies, which can affect model accuracy.

It includes:

 Handling missing values

 Removing noise and outliers

  Normalization and scaling

 Encoding categorical variables

 Data transformation

In brief:
Data pre-processing improves data quality and ensures better model performance.

19. Explain qualitative and quantitative data in details. Differentiate between the two.

Ans Qualitative and Quantitative Data

Data can be broadly classified into qualitative and quantitative based on the nature of information
they represent.

Qualitative Data

Qualitative data is descriptive and non-numerical in nature. It represents qualities, characteristics,

or categories of data rather than measurable quantities.

Characteristics

 Non-numeric

 Describes attributes or qualities

 Cannot be measured mathematically

 Often subjective

Types of Qualitative Data

1. Nominal Data – Categories with no natural order

Example: Gender, blood group, colors

2. Ordinal Data – Categories with a meaningful order

Example: Ratings (good, better, best), education level

Examples

 Eye color

 Type of vehicle

 Customer feedback (good/bad)

 Marital status

Uses

 Used in surveys, opinion polls, and classification problems

 Common in social sciences and market research

Quantitative Data
Quantitative data is numerical in nature and represents measurable quantities. It can be analyzed
using mathematical and statistical methods.

Characteristics

 Numeric

 Measurable

 Objective

 Can be statistically analyzed

Types of Quantitative Data

1. Discrete Data – Countable values

Example: Number of students, number of cars

2. Continuous Data – Measurable values within a range

Example: Height, weight, temperature

Examples

 Age

 Salary

 Distance

 Marks obtained

Uses

 Used in calculations, predictions, and regression analysis

 Common in scientific research and data analysis

Difference Between Qualitative and Quantitative Data

Aspect Qualitative Data Quantitative Data

Nature Descriptive Numerical

Measurement Cannot be measured numerically Can be measured numerically

Analysis Thematic or categorical Statistical and mathematical

Types Nominal, Ordinal Discrete, Continuous

Examples Gender, color, feedback Age, height, income

Usage in ML Classification tasks Regression and prediction

In Short

 Qualitative data describes qualities or categories.

  Quantitative data describes numbers and measurements.

20. What are the different causes of data issues in machine learning? What are the fallouts?

Ans Causes of Data Issues in Machine Learning and Their Fallouts

In machine learning, the quality of data directly affects model performance. Poor or problematic
data leads to inaccurate and unreliable models.

Causes of Data Issues

1. Missing Data
Occurs due to data entry errors, system failures, or incomplete records.

2. Noisy Data
Data contains random errors or incorrect values caused by faulty sensors or human mistakes.

3. Outliers
Extreme values that differ significantly from most data points.

4. Inconsistent Data
Conflicting information in different sources (e.g., different formats or units).

5. Duplicate Data
Repeated records due to data integration from multiple sources.

6. Imbalanced Data
One class has significantly more samples than others.

7. Irrelevant or Redundant Features

Features that do not contribute to prediction or are highly correlated.

8. Biased Data
Data does not represent real-world diversity.

9. Poor Data Labeling

Incorrect or inconsistent labels in supervised learning.

Fallouts of Data Issues

1. Low Model Accuracy

The model makes incorrect predictions.

2. Overfitting or Underfitting
The model learns noise or fails to learn meaningful patterns.

3. Biased Predictions
Unfair or unethical decisions.

4. Poor Generalization
Model performs well on training data but poorly on new data.
 5. Increased Training Time
More computational resources required.

6. Unreliable Decision Making

Leads to wrong business or scientific conclusions.

7. Model Failure in Real-world Use

The system may break or behave unexpectedly.

In Short

 Causes: Missing, noisy, biased, imbalanced, or inconsistent data

 Fallouts: Inaccurate, biased, and unreliable machine learning models

Below are clear, exam-ready answers to all three questions:

21. Basic Data Types in Machine Learning (with Examples)

1. Numerical (Quantitative) Data

Data represented by numbers and measurable values.
Example: Age = 25, Salary = ₹40,000, Temperature = 32°C

2. Categorical (Qualitative) Data

Data represented by categories or labels.
Example: Gender (Male/Female), Color (Red/Blue), City name

3. Ordinal Data
Categorical data with a meaningful order.
Example: Rating (Low, Medium, High), Education level

4. Binary Data
Data with only two possible values.
Example: Yes/No, 0/1, Pass/Fail

5. Time-Series Data
Data collected over time intervals.
Example: Daily stock prices, monthly sales

22. Differentiate

1. Categorical vs. Numeric Attribute

Aspect Categorical Attribute Numeric Attribute

Nature Non-numerical Numerical

Values Categories or labels Numbers

Operations Cannot perform arithmetic Arithmetic operations possible

Aspect Categorical Attribute Numeric Attribute

Examples Gender, Color, City Age, Height, Income

ML Usage Classification Regression

2. Dimensionality Reduction vs. Feature Selection

Aspect Dimensionality Reduction Feature Selection

Meaning Reduces features by creating new ones Selects a subset of existing features

Original Features Transformed Retained

Data Loss Possible Minimal

Techniques PCA, LDA, Autoencoders Filter, Wrapper, Embedded methods

Use Case High-dimensional data Feature relevance improvement

23. Main Activities Before Starting Modelling in Machine Learning

Main Activities Before Starting Modelling in Machine Learning

Before building a machine learning model, several key activities ensure the model is effective and
accurate. These activities are part of the data preparation and understanding process:

1. Understanding the Problem

o Define the objective of the ML task (classification, regression, clustering, etc.).

o Understand the expected outcomes and success criteria.

2. Data Collection

o Gather relevant data from databases, files, APIs, sensors, or online sources.

o Ensure the data is sufficient and representative of the problem.

3. Exploratory Data Analysis (EDA)

o Analyze the data to find patterns, trends, and anomalies.

o Use statistical summaries, charts, and visualizations.

4. Data Cleaning

o Handle missing values by imputation or removal.

o Remove duplicates and correct inconsistent entries.

o Handle noisy data and outliers.

5. Data Pre-processing
 o Normalize or scale numerical features.

o Encode categorical variables (e.g., one-hot encoding).

o Transform data if necessary (e.g., log transformation).

6. Feature Selection / Feature Engineering

o Select relevant features that contribute most to predictions.

o Create new features if needed to improve model performance.

7. Splitting the Dataset

o Divide the data into training, validation, and testing sets.

o Ensures the model can generalize to unseen data.

8. Understanding Data Biases and Imbalances

o Check for imbalanced classes and biases that may affect the model.

o Apply techniques like oversampling, undersampling, or weighting if needed.

In short:
Proper preparation before modelling ensures that the data is clean, relevant, and structured,
which is essential for building accurate, reliable, and efficient machine learning models.

unit 2

Q1 What is a model in context of machine learning? How can you train a

model?

ans **Model:**

A **machine learning model** is a mathematical representation or function that learns patterns,

relationships, and trends from given data. Once trained, the model can make predictions,
classifications, or decisions on new and unseen data based on what it has learned during training.

**Training a Model:**

1. **Collect Dataset:** Gather relevant and sufficient data related to the problem.

2. **Preprocess Data:** Clean the data by handling missing values, removing noise, and scaling
features if required.

3. **Split the Data:** Divide the dataset into training and testing (and sometimes validation) sets.

4. **Select an Algorithm:** Choose a suitable machine learning algorithm such as Linear

Regression, Decision Tree, or KNN.
5. **Train the Model:** Feed the training data into the algorithm so it can learn patterns.

6. **Adjust Parameters:** Optimize model parameters to minimize errors using a loss function.

7. **Evaluate the Model:** Test the trained model on unseen test data to check accuracy and
performance.

8. **Improve the Model:** If performance is low, repeat the process by tuning parameters or
improving data quality.

Q2 Explain, in details, the process of K-fold cross-validation.

ans ### K-Fold Cross-Validation (Detailed)

1. The complete dataset is divided into **K equal-sized parts**, called **folds**.

2. In the first iteration, **one fold is selected as the test set** and the remaining **K−1 folds are
used for training** the model.

3. The model is trained on the training folds and evaluated on the test fold.

4. This process is **repeated K times**, each time using a different fold as the test set.

5. Thus, **every data point is used once for testing and K−1 times for training**.

6. The performance scores (accuracy, error, etc.) from all K iterations are recorded.

7. The **final model performance** is calculated by taking the **average of all K results**.

**Benefits:**

* Provides a **more reliable and accurate estimate** of model performance.

* Makes **efficient use of the entire dataset**.

* Helps in **reducing overfitting**.

* Suitable for both small and large datasets.

Q3 Explain the bootstrap sampling. Why is it needed?

ans Bootstrap Sampling and Its Need

Bootstrap Sampling:

Bootstrap sampling is a resampling technique in which random samples are drawn from the
original dataset with replacement. Because sampling is done with replacement, the same data
point can appear multiple times in a single sample, while some data points may not appear at all.
Each bootstrap sample is usually of the same size as the original dataset.

Why Bootstrap Sampling is Needed:

It is very useful when the available dataset is small or limited.

It helps in estimating the accuracy and stability of a machine learning model.

It allows creation of multiple different training datasets from a single dataset.

It is widely used in ensemble learning methods like Bagging (Bootstrap Aggregating).

It helps in reducing variance and improving model performance.

Q4 What is the main purpose of a descriptive model? State some real-world problems solved using
descriptive models.

ans ### Purpose of Descriptive Model with Examples

**Purpose:**

Descriptive models are used to **analyze and describe patterns, relationships, trends, or
structures** present in a dataset. These models help in **understanding the data** rather than
predicting future outcomes. They summarize large and complex datasets into meaningful
information and provide insights that support decision-making.

**Key Points:**

* Focus on **data exploration and interpretation**

* Do **not predict future values**

* Often use unsupervised learning techniques

**Real-World Problems Solved Using Descriptive Models:**

1. **Customer Segmentation:** Grouping customers based on buying behavior or preferences.

2. **Market Basket Analysis:** Finding relationships between products frequently bought

together.

3. **Social Network Analysis:** Identifying communities or connections between users.

4. **Clustering Students Based on Performance:** Grouping students into categories such as high,
average, and low performers.

Q5 Explain the process of evaluating a linear regression model.

ans ### Purpose of Descriptive Model with Examples

**Purpose:**
Descriptive models are used to **analyze and describe patterns, relationships, trends, or
structures** present in a dataset. These models help in **understanding the data** rather than
predicting future outcomes. They summarize large and complex datasets into meaningful
information and provide insights that support decision-making.

**Key Points:**

* Focus on **data exploration and interpretation**

* Do **not predict future values**

* Often use unsupervised learning techniques

**Real-World Problems Solved Using Descriptive Models:**

1. **Customer Segmentation:** Grouping customers based on buying behavior or preferences.

2. **Market Basket Analysis:** Finding relationships between products frequently bought

together.

3. **Social Network Analysis:** Identifying communities or connections between users.

4. **Clustering Students Based on Performance:** Grouping students into categories such as high,
average, and low performers.

Q6 Differentiate (any two):

1. Predictive vs. descriptive models

2. Model underfitting vs. overfitting

3. Cross-validation vs. bootstrapping

ans

Predictive Models Descriptive Models

Used to predict future outcomes based on

Used to explain and understand data patterns
historical data, such as sales forecasting or price
and relationships present in the dataset.
prediction.

Work on unlabeled data where no predefined

Work on labeled data where the output variable
output is available, focusing on data
is already known during training.
exploration.
Underfitting Overfitting

Occurs when the model is too simple to learn the Occurs when the model is too complex and
underlying patterns in the data. It shows high bias learns noise along with patterns. It shows high
and gives poor performance on both training and variance and performs poorly on new or
testing data. unseen data.

The model memorizes training data instead of

The model fails to capture important relationships
generalizing, leading to poor generalization
among features, leading to low accuracy.
ability.

Cross-Validation Bootstrapping

The dataset is divided into fixed parts without

Samples are drawn with replacement, so the
replacement, and each part is used once for
same data point can appear multiple times in a
testing while the remaining data is used for
sample.
training.

Mainly used for model validation and performance Mainly used for resampling and creating
evaluation to estimate how well a model will multiple datasets, often to improve model
generalize to unseen data. stability and reduce variance.

Q7What is a target function? Express target function in context of a reallife example. How is the
fitness of a target function measured?

Ans Target Function (5 Marks)

A target function in machine learning is a mathematical function that defines the relationship
between input features and the output value that the model is trying to learn. The main goal of a
learning algorithm is to approximate this target function as accurately as possible.

Example:
In a house price prediction problem, the target function can be written as:
House Price = f(size, location, number of rooms)
Here, size, location, and rooms are input features, and house price is the output.

Fitness Measurement of Target Function:

The fitness of a target function is measured by evaluating how well the model performs on data.
This is done using:

1. Error metrics such as Mean Squared Error (MSE) or Accuracy.

2. Loss function value, which shows how far predictions are from actual values.

3. Validation performance, which checks model accuracy on unseen data to ensure good
generalization.

Q8 What are predictive models? What are descriptive models? Give examples of both types of
models. Explain the difference between these types of models.
Ans Predictive Models:
Predictive models are used to predict future or unknown values based on past data. These models
learn the relationship between input and output variables using labeled data. They are mainly
used for forecasting and decision-making.

Examples:

 Weather prediction

 Sales forecasting

 Stock price prediction

Descriptive Models:
Descriptive models are used to understand, analyze, and summarize data. They identify patterns,
groups, or relationships in data but do not predict future outcomes. These models usually work on
unlabeled data.

Examples:

 Customer clustering

 Topic modeling

 Market basket analysis

Difference:
Predictive models focus on prediction of future outcomes, while descriptive models focus on
understanding patterns and structure in existing data.

Q9 What is underfitting in context of machine learning models? What is the major cause of
underfitting?

Ans Underfitting in Machine Learning

Underfitting occurs when a machine learning model is too simple to capture the important
patterns, trends, and relationships present in the training data. Such a model is unable to learn
from the data properly and therefore gives poor performance on both training and test datasets.
Underfitting leads to inaccurate predictions and low overall model accuracy because the model
does not represent the data well.

Major Causes of Underfitting:

1. Very simple model selection, such as using a linear model for a complex problem.

2. Insufficient training time, where the model is not allowed to learn enough.

3. Poor feature selection, where important features are missing or ignored.

4. High bias, meaning the model makes strong assumptions about the data.

Q10 What is overfitting? When does it happen?

Ans Overfitting in Machine Learning

Overfitting occurs when a machine learning model learns the training data in too much detail,
including noise, outliers, and random variations, instead of learning the true underlying patterns.
Because of this, the model shows very high accuracy on training data but performs poorly on
unseen or test data. This means the model fails to generalize well to new data.

When Does Overfitting Happen?

1. When the model is too complex, such as having too many layers or parameters.

2. When the number of features is very high compared to the dataset size.

3. When the training data is limited or small.

4. When the model is trained for too many iterations without proper control.

5. When regularization or validation techniques (like cross-validation) are not used.

Q11 Explain bias-variance trade-off in context of model fitting

Ans Bias–Variance Trade-off in Model Fitting

The bias–variance trade-off explains the balance between two types of errors that affect the
performance of a machine learning model: bias and variance. The goal of model fitting is to
achieve a proper balance between them to get good accuracy on both training and unseen data.

Bias refers to the error caused by overly simple models that make strong assumptions about the
data. High bias models fail to capture important patterns, leading to underfitting and poor
performance on both training and test data.

Variance refers to the error caused by overly complex models that are very sensitive to small
changes in the training data. High variance models learn noise along with patterns, leading to
overfitting and poor performance on new data.

The trade-off means that reducing bias usually increases variance, and reducing variance usually
increases bias. A good model finds the right balance, minimizing total error and providing good
generalization on unseen data.

Q12 Write short notes on any two:

1. Holdout method

2. 10-fold cross-validation

3. Parameter tuning

Ans (a) Holdout Method

The holdout method is a simple technique for evaluating a machine learning model. In this
method, the dataset is divided into two parts: a training set and a testing set. The model is trained
on the training set and then evaluated on the testing set to check its performance.

This method is fast and easy to implement, making it suitable for large datasets. However, it can be
less reliable for small datasets because the evaluation depends heavily on how the data is split.
The performance may vary if a different split is used, leading to inconsistent results.

(b) 10-Fold Cross-Validation

10-Fold Cross-Validation is a more robust evaluation technique. In this method, the dataset is
divided into 10 equal parts (folds). The model is trained on 9 folds and tested on the remaining
fold. This process is repeated 10 times, with each fold used once as the test set.

The final performance is calculated as the average of all 10 results, making it more accurate than
the holdout method. It is especially useful for small datasets because it ensures that every data
point is used for both training and testing, reducing bias and variance in model evaluation.

(c) Parameter Tuning

Parameter tuning involves adjusting the hyperparameters of a machine learning model to improve
its performance. Hyperparameters are settings that control the learning process, such as learning
rate, number of trees in a random forest, or the number of neighbors in KNN.

Tuning is typically done using methods like Grid Search or Random Search, which systematically
test different combinations of hyperparameters. Proper parameter tuning can significantly improve
model accuracy, prevent overfitting, and help the model generalize better on unseen data.

Q13 Write the difference between Bagging vs. Boosting

Ans Difference Between Bagging and Boosting

Bagging Boosting

Stands for Bootstrap Aggregating. It builds

Builds models sequentially, where each model
multiple models in parallel using different subsets
tries to correct the errors of the previous one.
of the data.

Aims to reduce bias and improve prediction

Aims to reduce variance and prevent overfitting.
accuracy.

Models are dependent, and more weight is

Each model is trained independently and equally
given to models that perform better on difficult
weighted in the final prediction.
examples.

Works best when the base model is unstable (e.g., Works best with weak learners to convert them
decision trees). into a strong learner.
Bagging Boosting

Example: Random Forest Example: AdaBoost, Gradient Boosting

Summary: Bagging focuses on parallel model training to reduce variance, while Boosting focuses
on sequential training to reduce bias and improve accuracy.

Q14 Write the difference between Lazy vs. Eager learner

Ans Difference Between Lazy vs. Eager Learner

Lazy Learner Eager Learner

Learns only at the time of prediction. It stores the Learns a general model during training and
training data and waits until a query is made. uses it to make predictions later.

Training is fast, but prediction can be slow because Training is slower, but prediction is fast
computation happens at query time. because the model is already built.

Often used for instance-based learning. Often used for model-based learning.

Example: K-Nearest Neighbors (KNN) Example: Decision Trees, Linear Regression

Works well with small datasets but can be slow for Scales better for large datasets and can
large datasets. generalize well.

Summary: Lazy learners delay learning until prediction, while eager learners build a model in
advance to make faster predictions.

Q15 What is a feature? Explain with an example

Ans Feature in Machine Learning

A feature is an individual measurable property, characteristic, or attribute of the data that is used
as an input for a machine learning model. Features play a key role in helping the model understand
patterns, relationships, and trends in the data so it can make accurate predictions or classifications.
Choosing the right features and representing them properly is essential for building an effective
model.

Example:
In a house price prediction problem, the input features could include:

 Size of the house (in sq. ft.) – larger houses usually cost more.

 Number of bedrooms – more bedrooms can increase the price.

 Location of the house – houses in prime areas are more expensive.

 Age of the house – newer houses might have higher value.

Each of these properties is a feature that provides information to the model about factors affecting
the house price. The model uses these features to learn patterns and predict the price of new
houses. Including relevant and meaningful features improves model accuracy, while irrelevant
features can reduce performance.

Q16 Explain the process of encoding nominal variables.

Ans Encoding Nominal Variables in Machine Learning

Nominal variables are categorical variables that represent names, labels, or categories without any
intrinsic order. Machine learning models require numerical input, so encoding nominal variables is
the process of converting these categories into numbers.

Process of Encoding Nominal Variables:

1. Identify Nominal Variables:

Find features that are categorical without any order, e.g., Color = Red, Blue, Green or City =
Mumbai, Delhi, Bangalore.

2. Choose an Encoding Method:

o One-Hot Encoding: Creates a new binary column for each category.

Example: Color → Red(1,0,0), Blue(0,1,0), Green(0,0,1).

o Label Encoding: Assigns a unique integer to each category.

Example: City → Mumbai(0), Delhi(1), Bangalore(2).

3. Apply the Encoding:

Convert the categorical data into numeric form so it can be fed into machine learning
models.

4. Use Encoded Features in the Model:

Ensure the encoded features are properly used in training, avoiding any unintended ordinal
relationship if using label encoding on nominal data.

Benefits:

 Makes categorical data usable in models.

 Preserves all information from the original categories.

 Prevents errors caused by using raw text in computations.

Example:
Original data:

Color

Red

Blue

Green
Color

After One-Hot Encoding:

Red Blue Green

1 0 0

0 1 0

0 0 1

This numeric representation can now be used in any machine learning algorithm.

Q17 Explain the process of transforming numeric features to categorical features.

Ans Transforming Numeric Features to Categorical Features

Sometimes, numeric features need to be converted into categorical features to make them easier
to interpret or to improve model performance, especially when the numeric values fall into distinct
ranges or groups. This process is also called discretization or binning.

Transforming Numeric Features to Categorical Features

Process:

1. Identify numeric features – e.g., Age, Salary, Temperature.

2. Define bins – divide numbers into groups:

o Equal-width, equal-frequency, or custom bins.

3. Assign categories – replace numbers with labels.

Example: Age → Child (0–12), Teen (13–19), Adult (20–59), Senior (60+)

4. Use in model – categorical features can be used directly or encoded numerically.

Benefits:

 Simplifies numeric data

 Captures non-linear relationships

 Improves interpretability

 Example:

Age Age Group

8 Child

17 Teen
Age Age Group

25 Adult

65 Senior

Q18 Explain the wrapper approach of feature selection. What are the merits and de-merits of this
approach?

Ans Wrapper Approach of Feature Selection

The wrapper approach is a method of selecting the best subset of features by evaluating different
combinations of features using a machine learning model. It treats the feature selection process as
a search problem, where various subsets are tested, and the model’s performance determines
which features are selected.

Process:

1. Start with an empty set or all features.

2. Select a subset of features.

3. Train the model using this subset and evaluate performance.

4. Repeat with different subsets.

5. Choose the subset that gives the best model performance.

Merits:

 Often provides high accuracy because selection is based on actual model performance.

 Tailored to the specific model being used.

Demerits:

 Computationally expensive for large datasets with many features.

 Takes more time because multiple models need to be trained.

 Can overfit if not combined with proper validation.

This method is best for small to medium datasets where accuracy is more important than
computation time.

Q19 What is feature engineering? Explain, in details, the different aspects of feature engineering?

Ans Feature Engineering

Feature Engineering is the process of creating, modifying, and improving features in a dataset to
help machine learning models learn better and make more accurate predictions.

Key Aspects:
 1. Feature Creation: Generate new features from existing data.
Example: From Date of Birth, create Age.

2. Feature Transformation: Change scale or distribution to improve performance.

Example: Log-transform skewed income values.

3. Feature Scaling: Normalize or standardize features to a similar range.

Example: Scale height and weight between 0–1.

4. Feature Encoding: Convert categorical variables into numeric form.

Example: One-Hot Encoding for Color = Red, Blue, Green.

5. Feature Selection: Choose important features and remove irrelevant ones.

6. Handling Missing Values: Fill or remove missing data to improve quality.

7. Interaction Features: Combine variables to create meaningful features.

Example: BMI = Weight / (Height)^2.

Summary: Feature engineering transforms raw data into meaningful features, improving model
accuracy, interpretability, and training efficiency.

Q20 What is feature selection? Why is it needed? What are the different approaches of feature
selection?

Ans Feature Selection

Feature Selection is the process of choosing the most relevant features from a dataset that
contribute the most to a machine learning model’s accuracy, while removing irrelevant or
redundant features.

Why is it Needed?

1. Improves model accuracy by keeping only important features.

2. Reduces overfitting by eliminating noisy or irrelevant features.

3. Decreases computational cost and training time.

4. Makes the model simpler and easier to interpret.

Different Approaches of Feature Selection:

1. Filter Approach:

o Selects features based on statistical measures like correlation, chi-square, or

mutual information.

o Independent of any machine learning model.

2. Wrapper Approach:

o Evaluates different subsets of features using a specific model and selects the best-
performing subset.

3. Embedded Approach:
 o Feature selection occurs during model training, e.g., Lasso Regression or Decision
Trees automatically select important features.

Summary: Feature selection helps build efficient, accurate, and interpretable models by using only
the most useful features.

Q21 Explain the filter and wrapper approaches of feature selection. What are the merits and
demerits of these approaches?

Ans Filter and Wrapper Approaches of Feature Selection

1. Filter Approach:

 Features are selected based on statistical measures such as correlation, chi-square, or

mutual information.

 Independent of any machine learning model.

 Merits:

o Fast and computationally efficient.

o Works well with very large datasets.

 Demerits:

o May ignore interactions between features.

o Can select irrelevant features if statistical measures are misleading.

2. Wrapper Approach:

 Selects features by evaluating different subsets using a machine learning model and
choosing the subset that gives the best performance.

 Iteratively tests feature combinations with the model.

 Merits:

o Usually provides high accuracy because it is model-specific.

o Considers feature interactions.

 Demerits:

o Computationally expensive, especially for large datasets.

o Takes more time because multiple models are trained.

Summary:

 Filter: Fast, model-independent, but may miss important interactions.

 Wrapper: Accurate and considers interactions, but slow and resource-intensive.

Q22 Explain the overall process of feature selection

Ans Overall Process of Feature Selection

Feature selection is the process of identifying and keeping the most relevant features while
removing irrelevant or redundant ones to improve model performance. The overall process
involves the following steps:

1. Data Collection:

o Gather raw data from various sources that include all potential features.

2. Preprocessing:

o Clean the data by handling missing values, outliers, and noise.

o Convert categorical features into numeric form if needed.

3. Feature Evaluation:

o Assess the importance of each feature using statistical measures (Filter), model
performance (Wrapper), or embedded methods.

4. Feature Ranking or Scoring:

o Rank features based on their relevance or contribution to the model.

5. Feature Subset Selection:

o Choose the best subset of features that maximizes model performance while
reducing redundancy.

6. Model Training and Validation:

o Train the model using the selected features and evaluate performance on
validation/test data.

7. Iteration and Optimization:

o Refine the feature subset if needed to achieve better accuracy, reduce overfitting,
or improve interpretability.

Summary:
Feature selection ensures that the machine learning model is efficient, accurate, and easier to
interpret by focusing only on the most meaningful features.

Q23 Explain, with an example, the main underlying concept of feature extraction. What are the
most popular algorithms for feature extraction?

Ans Feature Extraction

Feature Extraction is the process of transforming the original features of a dataset into a new set of
features that are more informative and easier for a machine learning model to use. Unlike feature
selection, which picks existing features, feature extraction creates new features by combining or
reducing the original ones. This helps in reducing dimensionality and improving model
performance.

Example:
In an image recognition problem, an image may have thousands of pixel values as features.
Feature extraction can transform these into a smaller set of important features like edges, shapes,
or textures that capture the most relevant information. This reduces complexity while retaining
meaningful data for classification.

Popular Algorithms for Feature Extraction:

1. Principal Component Analysis (PCA): Reduces dimensionality by projecting data onto

principal components.

2. Linear Discriminant Analysis (LDA): Finds features that best separate classes.

3. Independent Component Analysis (ICA): Separates statistically independent features.

4. Autoencoders: Neural networks that learn compressed representations of input data.

Summary:
Feature extraction transforms raw data into informative, reduced, and meaningful features,
making models faster, simpler, and more accurate.