• Regularization eliminate or reduce the cofficeent of feature to zero if not needed. At a high level, L1 and L2 regularization are ways to “penalize” model complexity so your model doesn’t overfit the data.

🔹 Without regularization

• A model tries to fit the training data as closely as possible → can lead to:
  • Overfitting
  • Huge, unstable coefficients

🔹 With regularization

• We add a penalty for large coefficients, forcing the model to stay simpler.

L1 Regularization (Lasso)

• If a feature isn’t useful, eliminate it completely.

L2 Regularization (Ridge)

• “All features matter, but let’s tone them down.”