Generalized additive neural networks

Abstract

Generalized Additive Neural Networks (GANNs) integrate the power of neural networks and the interpretability of Generalized Additive Models (GAMs). This thesis try to offer a comprehensive theoretical context, starting with the origins of linear models, following with generalized models and ending with the explanation GANNs. We review the mathematical formulation, advantages, and disadvantages of GANNs compared to traditional models and some nonparametric methods. In addition, we will present a real application that demonstrate how GANNs have a good predictive accuracy and a great model interpretability. This thesis shows how flexible GANNs are,and how they could support machine learning and statistical modeling to improve.