Decision Trees Explained
Tree-based predictive models.
Structure
Nodes: conditions. Leaves: predictions. Branches: outcomes.
Splitting Criteria
Gini impurity: 1 - Σp². Entropy: -Σp log₂(p). Information gain: entropy before - weighted entropy after.
Pruning
Pre-pruning: max_depth, min_samples. Post-pruning: cost-complexity pruning.
Key Takeaways
- Interpretable tree structure
- Gini vs entropy for splitting
- Pruning controls complexity