Combined model of empirical study for credit risk management

In this paper, we studied the two most commonly used artificial intelligence methods (Multilayer Perceptron and Radial Basis Function network) to build the credit scoring model of applications, and analyzed the most important restraining factors of the applications of neural network which is the exponential increase in the variables bringing the model over-complex. On this basis, the author combines econometric analysis of the experience, through logistic regression the model can filter the variables with a high degree of correlation, which greatly reduces the complexity of the model, while the model has a better explanation, and thus improve the effect of neural network prediction models. The method can also be used for a variety of artificial intelligence applications to improve forecast model results.