The Study Of Personal Credit Scoring Model Based On Svm And Bp Neural Network

With the continuous improvement of living standards, the scale of personal consumption credit expanding. The credit loan that brought by this becomes key business of commercial bank. But at present our country’s credit system is not sound, credit records is almost blank, credit risk management is not mature. On this condition, most banks still use the traditional experience method. As the result, the risk of credit loan business increased. Many scholars put forwarded some scientific index systems and structured a lot of scoring models based on artificial intelligence and statistical methods. Though these methods improved the classification accuracy and robustness to a certain degree, it still can not meet the growing demand for credit business.On the basis of previous studies and combined with the actual situation in our country, many problems are pointed out in this paper. It includes unreasonable index systems and inefficiency scoring methods. In the light of existing problems in index system, we built an index system which is suited for our national conditions. It put forwards many opinions combined with experts’ advices. We also proposed a combined model which takes account of the high accuracy, high stability of BP neural network and the high efficiency of Support Vector Machine. The model has no local minimum value as well as SVM.Through empirical tests, the model proposed in this paper is better than BP neural network, SVM and Logistic—neural network on robustness and accuracy. So the improved BP neural network model that based on SVM is a good personal credit scoring model.