Python for Data Science — Your First Classification Model

In the previous article, we built our first Regression Model and learned how machine learning can predict numerical values such as:

• House Prices
• Revenue
• Sales
• Demand

But many real-world machine learning problems are different.

Instead of predicting numbers, we often need to predict categories.

For example:

• Spam or Not Spam
• Fraud or Not Fraud
• Customer Churn or No Churn
• Disease or No Disease

These are known as Classification Problems.

Classification is one of the most common applications of machine learning and powers many systems we interact with every day.

What Is Classification?

Classification is a type of supervised learning where the goal is to predict a category or class.

Unlike regression, which predicts numerical values:

House Price = $250,000

classification predicts labels such as:

Spam
Not Spam

or

Fraud
Not Fraud

The output is a category rather than a number.

Real-World Examples

Classification appears everywhere.

Email Filtering

Predict:

• Spam
• Not Spam

Fraud Detection

Predict:

• Fraudulent Transaction
• Legitimate Transaction

Customer Churn

Predict:

• Customer Leaves
• Customer Stays

Medical Diagnosis

Predict:

• Disease Present
• Disease Absent

Many machine learning systems are essentially classification systems.

Understanding Classes

A class is simply a possible outcome.

Example:

Customer Churn Dataset:

Possible classes:

Yes
No

The model learns patterns that help determine which class new observations belong to.

A Simple Example

Suppose we have customer data:

Features:

• Usage Hours
• Support Tickets

Label:

• Churn

The goal is to learn patterns connecting customer behavior to churn outcomes.

Our First Classification Algorithm

For beginners, one of the simplest classification models is:

Logistic Regression

The name can be confusing.

Despite the word “Regression,” it is actually a classification algorithm.

It is commonly used for predicting categories, especially binary outcomes.

Examples:

• Yes / No
• True / False
• Fraud / Not Fraud

Step 1: Define Features and Labels

Suppose:

X = data[["usage_hours", "support_tickets"]]

Features.

And:

y = data["churn"]

Label.

Step 2: Split the Data

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(
    X,
    y,
    test_size=0.2,
    random_state=42
)

As always:

Training data teaches the model.

Test data evaluates the model.

Step 3: Create the Model

from sklearn.linear_model import LogisticRegression

model = LogisticRegression()

At this point:

The model has not learned anything yet.

Step 4: Train the Model

model.fit(X_train, y_train)

The model studies historical examples and learns relationships between:

• Features
• Churn Outcomes

Step 5: Make Predictions

predictions = model.predict(X_test)

Now the model predicts:

Yes
No

for unseen customers.

This is classification in action.

What Is the Model Actually Learning?

Suppose historical data shows:

Get Sudha Rani Maddala’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

Customers who:

• use the product very little
• create many support tickets

are more likely to leave.

The model learns these patterns automatically.

It does not memorize rules.

It discovers relationships from data.

Why Classification Matters

Many business decisions involve categories.

Examples:

Banking

Approve Loan?

Yes / No

Marketing

Will Customer Respond?

Yes / No

Security

Is Transaction Fraudulent?

Yes / No

Healthcare

Disease Detected?

Yes / No

Classification models help automate these decisions.

Evaluating a Classification Model

In regression, we used metrics such as:

• Mean Absolute Error

Classification requires different metrics.

The simplest is:

Accuracy

from sklearn.metrics import accuracy_score

accuracy = accuracy_score(
    y_test,
    predictions
)

Accuracy measures:

What percentage of predictions were correct?

Example:

Actual:

Yes
No
Yes
No

Predicted:

Yes
No
No
No

Correct predictions:

3 out of 4

Accuracy:

75%

Is Accuracy Always Enough?

Not necessarily.

Imagine fraud detection.

Dataset:

99% Legitimate
1% Fraud

A model that predicts:

Always Legitimate

achieves:

99% Accuracy

But it completely fails at detecting fraud.

This is why data scientists use additional metrics.

We’ll explore those next.

Classification vs Regression

A useful comparison:

Both are supervised learning.

The difference lies in the type of prediction.

Common Beginner Mistakes

Mistake 1: Using Regression for Classification Problems

Always consider the type of output.

Mistake 2: Focusing Only on Accuracy

High accuracy does not always mean a useful model.

Mistake 3: Ignoring Class Imbalance

Some categories may be much rarer than others.

Mistake 4: Expecting Perfect Predictions

Classification models deal with probabilities and uncertainty.

Errors are inevitable.

Real-World Applications

Classification is used extensively in:

Fraud Detection

Identifying suspicious activity.

Healthcare

Diagnosing diseases.

Cybersecurity

Detecting malicious behavior.

Customer Analytics

Predicting churn and engagement.

Marketing

Targeting likely responders.

Many machine learning systems are fundamentally classification systems.

Thinking Like a Data Scientist

Beginners often ask:

Did the model predict correctly?

Experienced practitioners ask:

How well does the model perform across different types of errors?

That question becomes increasingly important as machine learning projects become more complex.

What’s Next?

We just introduced Accuracy.

But accuracy is only one way to evaluate classification models.

In many real-world situations, it can be misleading.

To evaluate models properly, we need to understand:

• Precision
• Recall
• F1-Score

These metrics help us understand not only whether predictions are correct, but also the types of mistakes the model makes.

Key Takeaway

Classification models predict categories rather than numerical values. They are widely used in applications such as fraud detection, spam filtering, customer churn prediction, and medical diagnosis.