Skip to content
Navigation
Dashboard
📖Reading•15 min

Standardization vs Normalization

Understand when to use standardization vs normalization for your data.

Standardization vs Normalization

Standardization (Z-Score)

x_std = (x - μ) / σ

  • Mean = 0, Std = 1
  • Use when: Data is normally distributed
  • Robust to outliers

    • Normalization (Min-Max)

    x_norm = (x - min) / (max - min)

  • Range: [0, 1]
  • Use when: Need bounded values
  • Sensitive to outliers

    • When to Use Which?

    Scenario | Recommendation |
    |----------|---------------|
    Neural Networks | Standardization |
    Image pixels | Normalization [0,1] |
    Tree-based models | Often not needed |
    Distance-based (KNN, SVM) | Standardization |
    Gradient descent | Always scale! |

    🎯 Key Takeaways

    • ✓Standardization centers data around zero
    • ✓Normalization bounds data to [0,1]
    • ✓Always scale for gradient-based methods
    • ✓Fit on training data, apply to test data

    📚 Additional Resources

    Previous
    Handling Data Imbalance
    Next
    Train/Validation/Test Splits