AdaBoost

Adaptive boosting that focuses on examples with large errors.

When to use:

• Have weak base learners
• Want classic ensemble method
• Simpler than gradient boosting

Strengths: Simple, less prone to overfitting, interpretable Weaknesses: Sensitive to noise and outliers, slower than modern boosting

Model Parameters

N Estimators (default: 50) Number of weak learners.

Learning Rate (default: 1.0) Shrinks the contribution of each learner.

Loss

• linear: Linear loss (default)
• square: Square loss
• exponential: Exponential loss

Random State (default: 42) Seed for reproducibility.