AdaBoost (Freund & Schapire), Breiman’s random forests (2001), and later gradient boosting (Friedman; XGBoost in 2014) showed that combining weak learners yields robust predictors.
On tabular data—credit, fraud, CTR—these methods beat poorly tuned nets. They remain mandatory baselines in many business problems.
Their success reinforced practices: careful validation, features, regularization. Deep learning did not erase them; it coexists depending on data type.
