Regularization

← Back to Model Evaluation

Techniques to prevent overfitting by adding constraints or penalties to the model. Essential for generalizing well to unseen data.

Types

• L1 (Lasso) — adds absolute value penalty, encourages sparsity
• L2 (Ridge) — adds squared penalty, shrinks weights
• Dropout — randomly zero neurons during training (neural networks)
• Early Stopping — stop training when validation loss stops improving
• Data Augmentation — artificially expand training data

Related

• Bias-Variance Tradeoff (regularization reduces variance)
• Linear Models (L1/L2 regularization)

---

ml regularization overfitting