Feature Scaling Explained Simply

What is Feature Scaling?

Feature scaling means changing numerical columns so their values are on a similar range or scale. This is useful when different features have very different units or value ranges.

Why Feature Scaling Matters

Some machine learning algorithms are sensitive to the size of numbers. If one feature has very large values, the model may give it too much influence even if it is not the most important feature.

• Helps models learn more effectively
• Prevents large-value features from dominating
• Improves performance for distance-based algorithms
• Helps gradient-based models train faster
• Makes features easier to compare

Business Example

Suppose we want to predict whether a customer will buy a product using:

Salary values are much larger than age values. Feature scaling helps place them on a more comparable scale before modelling.

Before and After Scaling

Common Scaling Methods

The two most common feature scaling methods are StandardScaler and MinMaxScaler.

StandardScaler

StandardScaler transforms data so that the feature has a mean of 0 and a standard deviation of 1.

StandardScaler is commonly used with algorithms such as logistic regression, support vector machines, K-nearest neighbours and neural networks.

MinMaxScaler

MinMaxScaler transforms values into a fixed range, usually between 0 and 1.

MinMaxScaler is useful when you want all features to stay within a clear minimum and maximum range.

Feature Scaling in Python: StandardScaler

Here is a simple example using Scikit-learn.

import pandas as pd
from sklearn.preprocessing import StandardScaler

data = {
    "Age": [25, 35, 45],
    "EstimatedSalary": [30000, 60000, 90000]
}

df = pd.DataFrame(data)

scaler = StandardScaler()

scaled_data = scaler.fit_transform(df)

print(scaled_data)

Feature Scaling in Python: MinMaxScaler

import pandas as pd
from sklearn.preprocessing import MinMaxScaler

data = {
    "Age": [25, 35, 45],
    "EstimatedSalary": [30000, 60000, 90000]
}

df = pd.DataFrame(data)

scaler = MinMaxScaler()

scaled_data = scaler.fit_transform(df)

print(scaled_data)

The transformed values will be between 0 and 1.

Correct Scaling Workflow

In machine learning, scaling should be done carefully to avoid data leakage.

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

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

scaler = StandardScaler()

X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)

Why Not Fit the Scaler on Test Data?

The test data should represent unseen future data. If we fit the scaler using test data, information from the test set leaks into training.

Which Models Need Scaling?

Tree-based models such as Decision Trees and Random Forests usually do not require feature scaling because they split data using thresholds.

Feature Scaling and Deep Learning

Neural networks usually perform better when input features are scaled. Large feature values can make training unstable or slow.

Common Beginner Mistakes

• Forgetting to scale numerical features
• Scaling before train-test split
• Using fit_transform on both training and testing data
• Scaling categorical values after encoding without understanding the effect
• Forgetting to save the scaler when deploying a model

Feature Scaling and Deployment

When using a trained model in real life, new data must be scaled in the same way as the training data. This means the scaler should be saved with the model.

import pickle

pickle.dump(scaler, open("scaler.pkl", "wb"))

Later, the same scaler can be loaded and used for new predictions.

scaler = pickle.load(open("scaler.pkl", "rb"))

new_data_scaled = scaler.transform(new_data)

Quick Practice

You are building a model using the following features:

Question: Should you consider feature scaling?

Suggested answer: Yes. The features have very different ranges, so scaling can help many machine learning models learn more effectively.

Key Takeaway

Feature scaling puts numerical features onto a similar scale. It is especially important for models that use distances, gradients or numerical optimisation, including logistic regression, KNN, SVM and neural networks.