Classification and Regression Trees

Summary

Establishes the CART methodology: growing decision trees by recursively choosing the split that most improves node purity, then pruning back the fully grown tree using held-out data.

Why it matters

It turned decision trees into a disciplined statistical method rather than an ad hoc heuristic, and the grow-then-prune strategy it introduced is still standard. Every tree ensemble in use today, random forests and gradient boosting alike, builds on the base learner formalized here.