Commerciral Bank Customers’ Credit Evaluation Model Study

The non-performing loan ratio of China’s commercial banks continued to rise. Thereis need to control individuals credit risk. The traditional manual credit check mode takes more time and has low accuracy. It is important to build personal credit evaluation modelfor higher accuracy and effiency.Domestic and foreign scholars have done a lot of research on credit evaluation model. The author found that numerous credit evaluation model selected variables and parameters set with strong subjective thought reviewing the existing literature. Besides, resarchers are lack of attention to accurately assess customer default rates. the author selected three kinds of single model: logistic regression, support vector machines and artificial neural network for the processing of personal credit assessment based on the customer data of German bank. The author introduced categorical variables in the process of logistic regression modeling. The cutoff value is set to 0.551 according to previous studies. The overall prediction accuracy rate is 62.5% andthe default customer identification rate of 44%. for the SVM model, the prediction accuracy of different kernel functions is compared to selected a best kernel function. It is provedthat polynomial kernel has a better effect rather than the theoretically recommended radial basis function.Generally, BP neural network model contains only one hidden layer. The paper compares one single hidden layer model with different nodes and the double hidden layer model.Single hidden layer model is selected as the most suitable one.By comparingthe results of simple models, it is obtained that neural network model has a higher recognition rate for breach of contract customersand a satistified overall prediction accuracy. Genetic algorithm is used to optimize the initial weights and threshold of neural network model to improve model stability. Logistic regression is introduced to reduce the impact of indicator collinearity.GA_BP model and Logit_GA_BP model are established by introduce GA and logistic model. The overall accuracy rate of the former combined model is66.17% for average total prediction and63.1% for default customers. The second model has an overall average prediction accuracy rate of 64.81% and default customers’ recognition rate of 68.32%.Logit_GA_BP model has higher prediction accuracy and stability of the strong for breach of default customers. It alsoclearly explain the influence of different variables on the credit status. There is much for reference of this method for commercial banks to establish an automated credit assessment system.