The Wisdom of Many Trees: A Narrative Review of Random Forests, from Randomized Partitions to the Ensemble's Theory
Year: 2026
Type: review
Abstract: Random forests are ensemble learning's most durable construction: hundreds of decorrelated decision trees, each grown on a bootstrap sample and randomized at every split, voting or averaging their way to predictions that outperform almost any single model. This article presents a narrative review of the primary literature of that construction. The synthesis is organized around three themes: the origins, in which Quinlan's and Breiman's decision trees, Breiman's bagging, and the randomized-tree designs of Ho and of Amit and Geman assembled the ensemble's ingredients; the canonical algorithm, in which Breiman's 2001 random forest fixed the mechanism---bootstrap aggregation with random feature selection at splits, out-of-bag error, and variable importance---and the experimental and software literature of Dietterich, Liaw and Wiener, and Oshiro and colleagues established its practice; and the interpretation and theory, in which Strobl's importance bias, Genner's variable selection, and Biau and Scornet's guided tour confronted the forest's opacity and built its mathematics. It is concluded that the random forest's success is a design lesson: that randomization, ensembled, buys both accuracy and robustness at the price of a theory the field is still completing.
Authors: Revista, Zen, IA, 10
Institution: Faculdades Guarulhos
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204785306
