Research on commercial bank credit risk evaluation model based on the integration of the probability distribution theory and the BP neural network technology

Studies have shown that BP (Back-Propagation) Neural Network can improve the accuracy of classification. However, most of the current research methods based on the expert-referendum integration strategy are all strongly subjective, which has a great uncertainty upon the credit risk evaluation results. To address this issue, the paper introduces the concept of credit risk coefficient and establishes an integral evaluation model based on the probability distribution theory and the BP neural network technology, which includes nine factor inputs and a three-tier structure to measure the credit risk output of commercial banks. The model can not only handle the nonlinear problems with good generalization ability, but also has a strong adaptive, self-organizing, and self-learning and regulating capability. The model also considers the preliminary classification results of the probability distribution and impact of each preliminary classification on the final decision, and employs Matlab7.0 to conduct an empirical test of the 161 sample data of a commercial bank. Research has shown that the model training results constructed by this paper has a smaller network prediction error and higher prediction accuracy than other classification methods. This model has been proved objective and valid, and has provided a theoretical and empirical basis for the bank to establish reliable and efficient evaluation system.