Artificial neural network and Bayesian network models for credit risk prediction

Abstract

Credit risk threatens financial institutions and may result in irrecoverable consequences. Tools for risk prediction can be used to reduce bank insolvency. This study compares Bayesian networks with artificial neural networks (ANNs) for predicting recovered value in a credit operation. The credit scoring problem is typically been approached as a supervised classification problem in machine learning. The present study explores this problem and finds that ANNs are a more efficient tool for predicting credit risk than the a¨ıve bayesian (NB) approach. The most crucial point is related to lending decisions, and a significant credit operation is associated with a set of factors to the degree that probabilities are used to classify new applicants based on their characteristics. The optimum achievement was obtained when the linear regression was equivalent to 0.2, with a mean accuracy of 85%. For the na¨ıve Bayes approach, the algorithm was applied to four datasets in a single process before the entire dataset was used to create a confusion matrix.