Explainable Boosting Machine

A glass-box model using cyclic gradient boosting on Generalized Additive Models (GAMs): it delivers high accuracy while providing exact, feature-level explanations.

The Explainable Boosting Machine (EBM) is a powerful 'glass-box' model, combining the transparency of Generalized Additive Models (GAMs) with the performance of boosting. It employs a cyclic gradient boosting algorithm over shallow trees to learn nonlinear main effects and a select number of pairwise interactions. This structure ensures every feature's contribution is explicitly additive and visualizable (e.g., as a 1D curve), providing both global and local interpretability. EBMs are proven to match the predictive accuracy of complex black-box methods like Random Forests and XGBoost—for instance, achieving $\sim0.928$ AUROC on the Adult Income dataset—making them a superior choice for regulated, high-stakes applications in finance and healthcare.

https://interpret.ml/docs/ebm.html