Wednesday, July 29, 2026

Decision Trees - I

DECISION TREES SERIES • TOPIC 1

Introduction to Decision Trees

Decision Trees are among the most intuitive and widely used machine learning algorithms. Instead of relying on complex mathematical equations, they make predictions by asking a sequence of logical questions, much like the way humans make decisions.

1. Why Were Decision Trees Invented?

Imagine you are a bank manager deciding whether to approve a loan application.

For every applicant, you ask a series of questions such as:

  • Is the salary above ₹8 lakh per year?
  • Does the applicant already have outstanding loans?
  • Is the credit score greater than 750?
  • Has the applicant ever defaulted on a loan?

Based on the answers, you finally arrive at one of two decisions:

✔ APPROVE
✖ REJECT

Notice Something Interesting

You never solved a complicated mathematical equation. Instead, you simply asked one question after another until you reached a decision.

This Is Exactly How a Decision Tree Works

A Decision Tree repeatedly asks questions about the input data. Every answer leads to another question until the algorithm finally reaches a prediction.

2. Decision Trees Are Everywhere in Daily Life

Although Decision Trees are an important machine learning algorithm, the underlying idea is something we use every day without even realising it.

Example 1: Doctor Diagnosing a Patient

A doctor often reaches a diagnosis by asking a sequence of questions.

Fever? / \ Yes No | | Cough? Allergy? / \ / \ Yes No Yes No Flu Viral Allergy Healthy
Observation

The doctor is not solving equations. Instead, each answer determines the next question, ultimately leading to a diagnosis.

Example 2: Buying a Laptop

Imagine helping someone choose a new laptop.

Need Gaming? | Yes / No | Budget > ₹1 lakh? | Yes / No | | High-end Mid-range
Observation

Every answer narrows the available choices until the most suitable laptop is selected.

Example 3: Airport Security

Airport security also follows a logical decision-making process.

Passport Valid? | Yes / No | | Visa Valid? Reject | Yes / No | | Board Reject
Key Insight

The same decision-making strategy appears in medicine, banking, shopping, airport security, and countless other real-world situations.

Part 1 Summary

  • Decision Trees imitate the way humans make decisions.
  • They solve problems by asking a sequence of logical questions.
  • No complicated mathematical equations are required to understand the basic idea.
  • Many everyday decision-making processes naturally resemble Decision Trees.

3. What Exactly Is a Decision Tree?

Formal Definition

A Decision Tree is a supervised machine learning algorithm that predicts an outcome by recursively splitting a dataset into smaller subsets based on the values of its input features.

At first glance, this definition may seem complicated because it contains several technical terms such as supervised learning, recursive splitting, and features.

Don't worry—we'll study every one of these concepts in detail throughout this Decision Tree series.

A Simpler Way to Think About It

A Decision Tree is simply a
📋 Flowchart of Questions

where every answer leads to another question
until a final prediction is reached.

Instead of solving one large mathematical equation, the algorithm repeatedly asks simple questions such as:

  • Is the customer's income greater than ₹10 lakh?
  • Is the applicant older than 30 years?
  • Does the patient have a fever?
  • Has the customer purchased before?

Every answer reduces the number of possible outcomes until only one prediction remains.

4. Why Is It Called a "Tree"?

Consider the following simple diagram.

Root │ ───────────────────── │ │ Question A Question B / \ / \ Yes No Yes No │ │ │ │ Prediction Prediction Prediction Prediction

The structure begins from one starting point and then branches into multiple paths.

This branching pattern resembles the way a real tree grows from a trunk into several branches.

Main Parts of a Decision Tree

Tree Component Meaning
Root The starting point of the tree.
Branches Different paths based on answers to questions.
Leaves Final predictions produced by the tree.
Note

We will study each of these terms in much greater detail in Topic 2: Basic Terminology.

5. Decision Trees Are Supervised Learning Algorithms

A Decision Tree learns from labelled training data.

Each training example contains:

  • Input Features (information about the example)
  • Target Output (the correct answer)

Example Training Dataset

Age Income Bought Product?
25 ₹5 Lakh No
40 ₹12 Lakh Yes
31 ₹7 Lakh No
52 ₹18 Lakh Yes

Understanding the Dataset

  • Age and Income are the input features.
  • Bought Product? is the target (output).

The Decision Tree analyses thousands (or even millions) of such examples and automatically discovers useful decision rules.

Example of a Learned Rule

Income > ₹10 lakh ? │ Yes / No │ │ Buy Don't Buy
Important Point

The programmer does not manually write this rule. The Decision Tree learns it automatically by analysing the training data.

Part 2 Summary

  • A Decision Tree predicts outcomes by asking a sequence of questions.
  • Its branching structure resembles a real tree, giving the algorithm its name.
  • Decision Trees learn from labelled training data.
  • The algorithm automatically discovers useful decision rules instead of requiring them to be programmed manually.

6. The Two Major Types of Decision Trees

Decision Trees are incredibly versatile because they can solve two fundamentally different kinds of machine learning problems. The type of Decision Tree used depends entirely on the kind of output we want to predict.

Two Categories of Decision Trees

Type Predicts Output Type
Classification Tree Categories or Classes Discrete Labels
Regression Tree Continuous Numbers Numeric Values

A. Classification Trees

A Classification Tree predicts categories (classes) rather than numerical values.

The output always belongs to one of a predefined set of labels.

Examples

Input Predicted Output
Email Spam / Not Spam
Patient Sick / Healthy
Customer Buy / Not Buy
Loan Application Approve / Reject
Animal Image Cat / Dog
Key Point

The output is always a label, not a numerical value.

B. Regression Trees

A Regression Tree predicts continuous numerical values.

Instead of selecting a category, the algorithm estimates an actual number.

Examples

Input Predicted Output
House ₹72 Lakh
Weather Conditions 32.5°C
Employee Profile ₹18.4 Lakh Salary
Rain Clouds 165 mm Rainfall
Stock Market Data ₹2,145 Share Price
Key Point

A Regression Tree predicts numbers, not categories.

Connection with Earlier Regression Algorithms

In the previous tutorial, we studied several regression algorithms such as:

  • Linear Regression
  • Polynomial Regression
  • Ridge Regression
  • Lasso Regression
  • Elastic Net Regression

All of these algorithms predict continuous numerical values.

A Regression Tree solves the same kind of problem, but it does so in a completely different way.

Important Difference

Traditional regression algorithms usually fit a mathematical equation (such as a straight line or a curve).

A Regression Tree does not fit an equation. Instead, it repeatedly splits the dataset into smaller groups and predicts a value for each group.

Classification Tree vs Regression Tree

Feature Classification Tree Regression Tree
Predicts Categories Numbers
Output Examples Yes / No, Spam / Not Spam ₹72 Lakh, 35°C, 165 mm
Problem Type Classification Regression
Example Loan Approval House Price Prediction

Part 3 Summary

  • Decision Trees can solve both classification and regression problems.
  • Classification Trees predict labels or categories.
  • Regression Trees predict continuous numerical values.
  • Regression Trees solve the same type of problems as Linear Regression and other regression algorithms, but they learn through recursive data splitting instead of fitting mathematical equations.

7. How Does a Decision Tree Think?

Unlike humans, a Decision Tree does not "think" in the traditional sense. Instead, it repeatedly asks carefully chosen questions that divide the data into smaller and more similar groups.

Each question eliminates some possibilities and moves the algorithm closer to the correct prediction.

Example: Predicting Whether Someone Will Buy a Car

Suppose we have the following information about every customer:

Available Features

  • Age
  • Salary
  • Married
  • Has Children

During training, the Decision Tree analyses thousands of examples and may discover that the following sequence of questions provides good predictions.

Example Decision Tree

Age > 30 ? / \ Yes No | | Salary > ₹10 L? Salary > ₹10 L? | | Yes / No Yes / No | | | | Buy Car Don't Buy Car Don't Buy Buy

Notice that every answer determines the next question. Eventually, the tree reaches a final decision.

What Happens After Every Question?

Every question divides the dataset into smaller, more homogeneous groups.

A homogeneous group is one in which the data points become increasingly similar to one another.

Example

Suppose we first split customers based on:

Salary > ₹10 lakh ?

Customers earning above ₹10 lakh may have a much higher probability of purchasing a car than customers earning below that amount.

The algorithm continues splitting each group until the remaining customers become sufficiently similar.

8. What Makes Decision Trees Special?

One of the biggest strengths of Decision Trees is that they do not assume any particular mathematical relationship between the input variables and the output.

This is a major difference compared with algorithms such as Linear Regression and Logistic Regression.

Comparison with Linear Regression

Linear Regression Decision Tree
Fits a mathematical equation (usually a straight line). Learns a sequence of decision rules.
Assumes an approximately linear relationship. Can naturally model highly non-linear relationships.

Example of a Non-Linear Relationship

Imagine the following customer behaviour:

Income < ₹5 lakh │ Don't Buy ₹5–15 lakh │ Buy Above ₹15 lakh │ Don't Buy

This relationship is clearly not a straight line.

Buying behaviour increases initially, then decreases again for higher income groups.

Why Linear Models Struggle Here

A simple linear model tries to fit one mathematical relationship across the entire dataset.

When customer behaviour changes at different income ranges, one straight line cannot accurately represent every pattern.

Result

The predictions may become inaccurate because the relationship is highly non-linear.

Why Decision Trees Handle This Naturally

A Decision Tree is free to create different decision boundaries for different parts of the data.

Instead of fitting a single equation, it can create rules such as:

If Income < ₹5 lakh ↓ Don't Buy Else If Income ≤ ₹15 lakh ↓ Buy Else ↓ Don't Buy

This flexibility enables Decision Trees to model many complex, real-world relationships that are difficult for simple linear models to capture.

Part 4 Summary

  • Decision Trees make predictions by asking a sequence of questions.
  • Each question divides the data into smaller, more homogeneous groups.
  • Unlike Linear Regression, Decision Trees do not assume a straight-line relationship.
  • They naturally model complex, non-linear patterns by creating different decision rules for different regions of the data.

9. Where Are Decision Trees Used?

Decision Trees are one of the most widely used machine learning algorithms because they are easy to understand, require very little data preprocessing, and perform well across a wide variety of real-world applications.

From banking to healthcare, and from agriculture to cybersecurity, Decision Trees help organisations make accurate, explainable decisions based on historical data.

Common Applications of Decision Trees

🏦 Banking

  • Loan approval
  • Credit scoring
  • Customer risk analysis

🏥 Healthcare

  • Disease diagnosis
  • Treatment recommendations
  • Patient risk prediction

🏭 Manufacturing

  • Quality control
  • Defect detection
  • Predictive maintenance

🛒 Retail

  • Customer segmentation
  • Product recommendations
  • Demand forecasting

📢 Marketing

  • Campaign targeting
  • Lead scoring
  • Customer behaviour analysis

🔐 Cybersecurity

  • Intrusion detection
  • Malware classification
  • Fraud detection

👥 Human Resources

  • Employee attrition prediction
  • Recruitment screening
  • Performance analysis

🌾 Agriculture

  • Crop disease detection
  • Yield prediction
  • Irrigation planning

💹 Finance

  • Fraud detection
  • Investment risk assessment
  • Credit default prediction

10. Why Are Decision Trees So Popular?

Despite being one of the oldest machine learning algorithms, Decision Trees remain extremely popular because they combine simplicity, flexibility, and interpretability.

✅ Easy to Understand

Decision Trees resemble human decision-making and are therefore easy to explain, even to non-technical stakeholders.

📊 Handles Different Data Types

They work naturally with both numerical and categorical features.

⚙ Minimal Preprocessing

Decision Trees usually require little or no feature scaling, normalisation, or complex data transformations.

📈 Captures Non-Linear Patterns

Unlike simple linear models, Decision Trees naturally model complex relationships between variables.

🔍 Highly Interpretable

Every prediction can be traced through a sequence of logical questions, making the model transparent and explainable.

🌳 Foundation of Powerful Algorithms

Many state-of-the-art machine learning models are built by combining multiple Decision Trees.

Decision Trees Are the Building Blocks of Modern Machine Learning

Although a single Decision Tree is already useful, modern machine learning often combines many trees together to build even more powerful predictive models.

Algorithm Main Idea
Random Forest Trains many Decision Trees independently and combines their predictions.
Gradient Boosting Builds Decision Trees sequentially, with each tree correcting the mistakes of the previous one.
XGBoost / LightGBM / CatBoost Highly optimised boosting algorithms widely used in industry and machine learning competitions.
Important Note

Understanding a single Decision Tree thoroughly makes it much easier to learn advanced ensemble algorithms such as Random Forest and Gradient Boosting, which are among the most successful machine learning techniques used today.

Part 5 Summary

  • Decision Trees are widely used across banking, healthcare, manufacturing, retail, finance, cybersecurity, agriculture, and many other industries.
  • They are popular because they are easy to understand, require minimal preprocessing, and naturally handle complex, non-linear relationships.
  • Decision Trees are also the foundation of powerful ensemble algorithms such as Random Forest, Gradient Boosting, XGBoost, LightGBM, and CatBoost.

11. A Simple Worked Example

The best way to understand a Decision Tree is to see one learn from a small dataset.

Suppose we want to predict whether people will play cricket based only on the weather.

Training Dataset

The algorithm is provided with the following labelled examples.

Weather Play Cricket?
Sunny No
Sunny No
Rainy Yes
Cloudy Yes
Rainy Yes
Sunny No

What Does the Decision Tree Learn?

After examining the training data, the Decision Tree discovers that the Weather feature alone is sufficient to make accurate predictions.

It therefore learns a simple set of decision rules instead of trying to memorise every individual example.

Learned Decision Tree

Weather │ ┌────────┬──────────┐ │ │ │ Sunny Rainy Cloudy │ │ │ No Yes Yes
Observation

The Decision Tree has converted the training data into a simple, easy-to-understand flowchart of decisions.

Making Predictions

Once the tree has been trained, making predictions becomes extremely simple.

Input Decision Path Prediction
Sunny Weather → Sunny No
Rainy Weather → Rainy Yes
Cloudy Weather → Cloudy Yes

Why Does This Work?

During training, the algorithm searches for questions that best separate the training examples into groups containing similar outcomes.

In this small dataset:

  • Every Sunny example resulted in No.
  • Every Rainy example resulted in Yes.
  • Every Cloudy example also resulted in Yes.

Because the weather perfectly separates the data, the algorithm can make accurate predictions using just one question.

An Important Observation

This example is intentionally simple so that the underlying idea is easy to understand.

Real-world Decision Trees usually contain:

  • Many input features
  • Hundreds or even thousands of decision nodes
  • Much deeper tree structures
  • Far more complex decision rules

Nevertheless, the fundamental principle remains exactly the same: repeatedly ask questions that divide the data into increasingly homogeneous groups until a reliable prediction can be made.

Part 6 Summary

  • A Decision Tree learns patterns from labelled training data.
  • It converts those patterns into a sequence of easy-to-follow decision rules.
  • Predictions are made by following a path from the root node to a leaf node.
  • The same learning principle applies to both small examples and very large real-world datasets.

12. Decision Trees vs. Rule-Based Programming

One of the most common questions beginners ask is:

"Isn't a Decision Tree just a collection of if-else statements?"

The answer is:

Yes... and No.

Why the Answer is "Yes"

After a Decision Tree has been trained, every prediction can be represented as a sequence of nested if-else conditions.

Consider the following Decision Tree:

Income > ₹10 lakh? Yes │ Age > 30 ? Yes No │ │ Approve Reject │ No │ Reject

This tree can easily be converted into ordinary programming logic.

Equivalent Python if-else Logic

if income > 1000000:
    if age > 30:
        approve
    else:
        reject
else:
    reject

Therefore, the final model produced by a Decision Tree can indeed be written as a series of nested if-else statements.

Why the Answer is Also "No"

Although the finished Decision Tree resembles ordinary programming, the way those rules are created is completely different.

This is the fundamental difference between traditional programming and machine learning.

Traditional Programming vs. Decision Trees

Traditional Programming Decision Tree Learning
A programmer manually writes every rule. The algorithm automatically discovers rules from data.
Rules remain fixed unless the programmer changes them. Rules change automatically when new training data is used.
Human experience creates the logic. Historical data creates the logic.
Difficult to maintain when rules become numerous. Learns thousands of rules automatically if needed.

A Real-World Example

Imagine you are building a loan approval system.

Traditional Programming

  • You interview banking experts.
  • You manually write hundreds of rules.
  • Every policy change requires code changes.
  • The program cannot improve by itself.

Machine Learning

  • Collect thousands of previous loan applications.
  • Train a Decision Tree using historical outcomes.
  • The algorithm discovers useful decision rules automatically.
  • Retraining with new data allows the model to adapt.

The Key Difference

A Decision Tree is not special because it uses if-else logic.

The remarkable part is that it learns those rules automatically from training data instead of requiring a human programmer to write every rule manually.

This automatic learning capability is what makes Decision Trees a machine learning algorithm rather than a traditional computer program.

Part 7 Summary

  • The final Decision Tree can be represented as nested if-else statements.
  • The crucial difference lies in how those rules are created.
  • Traditional programs rely on manually written rules.
  • Decision Trees automatically discover rules by learning from labelled training data.
  • This ability to learn from data is what makes Decision Trees a true machine learning algorithm.

13. Advantages of Decision Trees (High-Level)

Decision Trees are among the most popular machine learning algorithms because they combine simplicity, interpretability, and flexibility.

Although we'll study these advantages in much greater detail later in this course, it's useful to understand the major benefits at a high level.

🌳 Easy to Visualize

A Decision Tree can be drawn as a flowchart, making it much easier to understand than many other machine learning models.

💬 Easy to Explain

Each prediction follows a sequence of logical questions, making Decision Trees highly interpretable for both technical and non-technical users.

🔢 Handles Mixed Data Types

Decision Trees work naturally with both numerical and categorical features without requiring special encoding in many situations.

⚙ Minimal Data Preprocessing

Unlike many algorithms, Decision Trees usually do not require feature scaling, normalization, or standardization.

📈 Learns Non-Linear Relationships

By repeatedly splitting the data, Decision Trees can model complex patterns that simple linear models cannot capture.

🤝 Captures Feature Interactions

Decision Trees naturally learn how combinations of multiple features influence the final prediction.

Why Are Decision Trees Considered Interpretable?

Many machine learning algorithms behave like "black boxes", meaning it is difficult to understand why they produced a particular prediction.

Decision Trees are different.

Every prediction follows a visible sequence of decisions:

Age > 30? │ Yes │ Salary > ₹10 lakh? │ Yes │ Approve Loan

Since every decision is visible, we can explain exactly why the model reached its conclusion.

14. Limitations of Decision Trees (High-Level)

Although Decision Trees are powerful and easy to understand, they are not perfect.

Understanding these limitations helps explain why advanced ensemble methods such as Random Forest and Gradient Boosting were developed.

⚠ Can Overfit the Training Data

A very deep tree may memorise the training examples instead of learning general patterns, resulting in poor performance on unseen data.

🔄 Sensitive to Small Data Changes

Even a small change in the training dataset can sometimes produce a noticeably different Decision Tree.

🌿 Deep Trees Become Harder to Interpret

Although small trees are easy to understand, very large trees may contain hundreds of decision nodes and become difficult to follow.

🎯 Greedy Splitting

Decision Trees usually choose the best split available at each step, but this does not always guarantee the globally optimal tree.

Don't Worry—We'll Study These in Detail

At this stage, you only need a high-level understanding of these limitations.

In later topics, we'll explore:

  • Why overfitting occurs.
  • How tree pruning helps reduce overfitting.
  • Why Random Forests are more stable than individual Decision Trees.
  • How Gradient Boosting improves predictive accuracy.

Advantages vs. Limitations

Advantages Limitations
Easy to understand and visualize Can overfit training data
Handles mixed feature types Sensitive to small dataset changes
Learns non-linear relationships Very deep trees become difficult to interpret
Requires little preprocessing Greedy splitting may not find the optimal tree

Part 8 Summary

  • Decision Trees are easy to understand, explain, and visualize.
  • They naturally handle numerical and categorical data and require very little preprocessing.
  • They can model complex, non-linear relationships and feature interactions.
  • However, they may overfit, become unstable with small data changes, and very deep trees can be difficult to interpret.
  • These limitations motivated the development of ensemble methods such as Random Forests and Gradient Boosting.

Frequently Asked Questions

These are some of the most common Decision Tree questions. 

1. What is a Decision Tree?

A Decision Tree is a supervised machine learning algorithm that makes predictions by recursively splitting data based on feature values, producing a tree-like structure of decision rules.

2. Is a Decision Tree Used for Classification or Regression?

Both.

  • Classification Trees predict categories or class labels.
  • Regression Trees predict continuous numerical values.

3. Why Is It Called a "Tree"?

The model starts from a single root node, branches into multiple decision paths, and finally ends at leaf nodes, making its structure resemble an upside-down tree.

4. Does a Decision Tree Require Feature Scaling?

No.

Unlike algorithms such as K-Nearest Neighbours (KNN) or Support Vector Machines (SVM), Decision Trees work directly with the original feature values.

Feature scaling, normalization, or standardization is generally unnecessary.

Quick Revision Table

Concept Key Idea
Decision Tree Learns by asking a sequence of questions.
Learning Type Supervised Learning
Main Structure Root → Branches → Leaf Nodes
Problems Solved Classification and Regression
Feature Scaling Not Required
Relationship Assumption No assumption of linearity
Biggest Strength Easy to interpret and explain

Topic Summary

By the end of this topic, you should understand the following core ideas:

  • A Decision Tree is a supervised machine learning algorithm.
  • It learns by asking a sequence of questions about the input data.
  • It can solve both classification and regression problems.
  • Its predictions are easy to understand because they follow a clear decision path.
  • Unlike Linear Regression, it does not assume a linear relationship between inputs and outputs.
  • It naturally models complex, non-linear relationships.
  • The learned model resembles a flowchart or a collection of nested if-else statements.
  • Decision Trees require very little data preprocessing.
  • They are the foundation of advanced ensemble algorithms such as Random Forest and Gradient Boosting.

Key Takeaways

  • Think of a Decision Tree as an intelligent flowchart.
  • Every internal node represents a question.
  • Each branch represents one possible answer.
  • Each leaf node contains the final prediction.
  • The algorithm learns these decision rules automatically from labelled training data.
  • Its simplicity and interpretability make it one of the most important algorithms in machine learning.

Next Topic

Topic 2: Basic Terminology

Before we can build or analyse Decision Trees, we need to understand the terminology used throughout the rest of the course.

In the next topic, we'll study the following concepts in detail:

  • Root Node
  • Decision Node
  • Leaf Node
  • Parent Node
  • Child Node
  • Branches
  • Subtrees
  • Tree Depth
  • Tree Height
  • Tree Levels
Why This Matters

These terms form the vocabulary of Decision Trees. We'll use them repeatedly in every subsequent topic, so understanding them now will make the remaining lessons much easier to follow.

🎉 End of Topic 1

You have now completed Topic 1: Introduction to Decision Trees.

In the next lesson, we'll begin building a solid foundation by learning the essential terminology used in every Decision Tree algorithm.

No comments:

Post a Comment

Decision Trees - II

Topic 2: Basic Terminology of Decision Trees This topic form...