Credit Risk Classification Using Discriminative Restricted Boltzmann Machines

Abstract

Credit risk analysis plays an important role in the financial market. In this paper, discriminative restricted Boltzmann machine (RBM) is used in credit risk classification. RBM is a generative model associated with an undirected graph, which can capture complicated features from observed data, and after introducing discriminative component into RBM, it can be used to train a non-linear classifier. The method is tested in a real-world credit risk prediction task, and the empirical results demonstrate the advantage of the method over other competing ones.