CatBoost
Train CatBoost to predict categorical outcomes
Yandex's gradient boosting that handles categorical features natively without encoding.
When to use:
- Have categorical features
- Don't want to manually encode categories
- Need robust performance with minimal tuning
- Unbalanced datasets
Strengths: Handles categories natively, less overfitting, good default parameters, ordered boosting Weaknesses: Slower than LightGBM, larger models
Model Parameters
Iterations (default: 100) Number of boosting iterations (like n_estimators).
Depth (default: 6) Tree depth.
Learning Rate (default: 0.03) Lower default than other boosting models.
L2 Leaf Reg (default: 3.0) L2 regularization coefficient.
Border Count (default: 254) Number of splits for numerical features.
Auto Class Weights Automatically handle class imbalance.
Random State (default: 42) Seed for reproducibility.