Leo Breiman (1928-2005) was a statistician at the University of California, Berkeley and a member of the National Academy of Sciences whose work helped turn statistics toward the data-driven, predictive style that became modern machine learning. His Berkeley page describes a focus on computationally intensive multivariate analysis, especially nonlinear methods for pattern recognition and prediction in high-dimensional spaces.
Breiman co-authored Classification and Regression Trees (CART), developing decision trees as computationally efficient alternatives to neural networks. He later invented two of the most important ensemble methods in machine learning: bagging, which averages many models trained on resampled data, and random forests, which combines many randomized decision trees into a single robust predictor. Random forests remain one of the most widely used and reliable off-the-shelf algorithms for tabular data.
In a famous 2001 essay, “Statistical Modeling: The Two Cultures,” Breiman argued that the field should pay far more attention to predictive accuracy and algorithmic models rather than insisting on simple, interpretable data-generating assumptions. That argument anticipated much of the debate that machine learning would later force on statistics.