Research On The Application Of PSO-BP Neural Network To Credit Risk Assessment For Commercial Banks

Credit risk is one of the most essential risks in financial institution. When solving the problem of credit risk assessment, neural network has unique advantages for its non-linear mapping ability. With BP neural network (BP-NN), it is possible to perfectly realize the non-linear mapping relationship between credit index and the classification of credit grade, thus achieving the function of classifying customer into different credit grades according credit index. However, the parameters of BP-NN are based on partial information in the whole parameter space rather than the global optimal value, which will definitely reduce the convergence rate and prediction accuracy of BP-NN. Consequently, this; dissertation is an attempt to study Particle Swarm Optimization (PSO) algorithm with the function of global optimization, for it can improve the learning strategy of BP-NN and fix the loophole of parameters setting in BP-NN.This dissertation mainly dealt with the following work. Firstly, on the basis of the development and present situation of credit risk assessment for commercial banks, the author made a general summary concerning the traditional and modern methods applied to credit risk assessment. Secondly, in light of a systemic research on the theoretical fundamentals of BP-NN and PSO algorithm, and with a view to solving the problems of poor partial searching capability and premature convergence in the standard PSO algorithm, this dissertation proposed a modified PSO algorithm, which was applied to optimize the weight value of BP-NN, and thus created a new BP-NN model based on the modified PSO algorithm. Finally, the new model has been used to conduct a case emulation involving the data of credit risk assessment from a commercial bank. In comparison with traditional BP-NN, GA-BP-NN and standard PSO-BP-NN model, the experimental results indicated that the new model has quicker convergence rate and higher prediction accuracy, thus proving that the new model is feasible and effective in the field of credit risk assessment for commercial banks.