Commercial Bank Credit Risk Evaluation Based On Neural Network And Evidence Theory

With increasing credit defaults, credit risk management has been an important measure for commercial banks to control the risks. In this paper, I give a brief introduction to the features of commercial bank risks and the techniques of evaluating them. I take neural network models as research priorities, and do the following three aspects:In the first place, we introduce BP,SVM and RBF networks, taking place of conventional expert systems. Conventional expert systems are subjective due to different professional level and concerns. The neural network has an advantage over anti-interference and dynamic adjustment, thus we could effectively reduce subjective probabilities by using different neural networks.. However, clients'credit evaluating indicators are substantial, which causes long training time problem, in order to resolve it, we pull in rough set theory to improve the integration efficiency. And it has been verified that the rough set could substantially reduce work of neural network training.In the next place, an improved D-S theory is introduced, and we spread it from two dimensions to multi-dimention, try to give a more precise judgement means. D-S theory is an effective data fusion method, which could integrate over-lapped related multi-source information without prior knowledge and conditions. However, conventional D-S theory didn't take theory validity and importance, besides, it is too rough to judge if the evidences are conflict. We take the datas processing by networks as fundamental probabilities, which could get rid of subjective interference.Again, we give an evaluation example for commercial bank credit risk, to verify the sufficience of our method. While processing datas, we take the average to determine conflict degrees among clients. And the result has demonstrated that our method is effective.Finally, we simply introduce the current status of credit risk means in our country, and give a possible study direction.