Odds ratio function estimation using a generalized additive neural network

In biomedical research, generalized artificial neural networks (GANNs) have been proposed as an alternative to a multi-layer perceptron owing to their greater ability to generate more interpretable results. GANNs were inspired by statistical generalized additive models (GAMs), and because of the parallelism that can be established between ANNs and GAMs, it is natural for advances in GAMs to be incorporated into the field of neural networks. A GANN with a flexible link function was recently proposed, with results similar to those of a GAM with the same type of link function. However, in the medical field, more improvements must be introduced to obtain even more interpretable, and consequently more useful, ANNs. In this study, an algorithm for estimating the odds ratio function for continuous covariates is proposed, which increases the interpretability of a GANN.