Feature Scaling: Normalization and Standardization

Feature Scaling

Many ML algorithms (e.g., K‑means, SVM) are sensitive to feature magnitude. Scaling brings features onto a comparable scale.

Normalization (min‑max scaling)

from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()
scaled = scaler.fit_transform(df[['price', 'quantity']])
scaled_df = pd.DataFrame(scaled, columns=['price_norm', 'quantity_norm'])

Standardization (zero‑mean, unit‑variance)

from sklearn.preprocessing import StandardScaler
std_scaler = StandardScaler()
std_scaled = std_scaler.fit_transform(df[['price', 'quantity']])
std_df = pd.DataFrame(std_scaled, columns=['price_std', 'quantity_std'])

When to use which?

• Normalization: when you need bounded values (0‑1), e.g., neural networks with sigmoid.
• Standardization: works well for algorithms assuming Gaussian distribution.

Scaling pipelines with Pipeline

from sklearn.pipeline import Pipeline
pipe = Pipeline([
    ('imputer', SimpleImputer(strategy='median')),
    ('scaler', StandardScaler())
])
X_scaled = pipe.fit_transform(df[numeric_features])

Visualizing before/after scaling

fig, ax = plt.subplots(1,2, figsize=(12,5))
ax[0].hist(df['price'], bins=30, color='skyblue')
ax[0].set_title('Original')
ax[1].hist(scaled_df['price_norm'], bins=30, color='orange')
ax[1].set_title('Normalized')
plt.show()