Leo Breiman

Created bagging and random forests, and formalized decision trees.

Biography

An American statistician who moved between academia and consulting before returning to Berkeley. He argued forcefully that predictive accuracy, assessed honestly out of sample, deserved standing alongside classical model-based inference.

Key contributions

• Co-authored Classification and Regression Trees (1984), formalizing tree induction and pruning.
• Introduced bagging, reducing variance by averaging models fitted to bootstrap resamples (1996).
• Developed random forests, decorrelating bagged trees by restricting features at each split.

Impact

Random forests remain one of the strongest general-purpose methods for tabular data, and the ensemble principle he established underlies gradient boosting as well.