Neural Network Metalearning for Credit Scoring

Abstract

In the field of credit risk analysis, the problem that we often encountered is to increase the model accuracy as possible using the limited data. In this study, we discuss the use of supervised neural networks as a metalearning technique to design a credit scoring system to solve this problem. First of all, a bagging sampling technique is used to generate different training sets to overcome data shortage problem. Based on the different training sets, the different neural network models with different initial conditions or training algorithms is then trained to formulate different credit scoring models, i.e., base models. Finally, a neural-network-based metamodel can be produced by learning from all base models so as to improve the reliability, i.e., predict defaults accurately. For illustration, a credit card application approval experiment is performed.