Functional ANOVA

Functional ANOVA (fANOVA) is a statistical technique used to decompose complex models into simpler, interpretable components, revealing how individual input variables and their interactions contribute to the model's output. Current research focuses on developing efficient algorithms, such as Explainable Boosting Machines (EBMs), GAMI-Net, and GAMI-Tree, to fit fANOVA models to various data types, including those generated by black-box machine learning models, and improving the accuracy and interpretability of these decompositions. This approach enhances the transparency and trustworthiness of machine learning models, particularly valuable in fields requiring accountability and decision explainability, such as medicine and finance. Furthermore, fANOVA is being extended to analyze hyperparameter importance in complex models like deep neural networks.