Classification and Regression Trees (CART)

Formalized binary recursive partitioning, Gini impurity splitting, and cost-complexity pruning for decision trees.

The core tree-growing engine utilized by Random Forests, Extra Trees, and Gradient Boosted Decision Trees.

Assumptions

• Complex multidimensional decision boundaries can be approximated by recursive orthogonal axis-aligned hyperplanes

Limitations

• Single decision trees suffer from high variance and instability; small perturbations in data alter entire tree structure