Credit risk assessment with a multistage neural network ensemble learning approach

Abstract

In this study, a multistage neural network ensemble learning model is proposed to evaluate credit risk at the measurement level. The proposed model consists of six stages. In the first stage, a bagging sampling approach is used to generate different training data subsets especially for data shortage. In the second stage, the different neural network models are created with different training subsets obtained from the previous stage. In the third stage, the generated neural network models are trained with different training datasets and accordingly the classification score and reliability value of neural classifier can be obtained. In the fourth stage, a decorrelation maximization algorithm is used to select the appropriate ensemble members. In the fifth stage, the reliability values of the selected neural network models (i.e., ensemble members) are scaled into a unit interval by logistic transformation. In the final stage, the selected neural network ensemble members are fused to obtain final classification result by means of reliability measurement. For illustration, two publicly available credit datasets are used to verify the effectiveness of the proposed multistage neural network ensemble model.