The Bp Neural Network Research And Its Application In Personal Credit Evaluation

With the rapid development of economy, credit consuming surface, and individual consumption loans, for example housing mortgages loan、 car loan、 credit card and so on, need credit guarantee, so personal credit assessment has an important significance for commercial banks to control the credit risk. Artificial Neural Network(ANN) is an intelligent information processing system of the brain-type, which simulates the structure and functions of the brain, the area of its application is expanding and the potential is increasingly obvious, and in the field of personal credit evaluation it also has a very good application prospect. The topic of this thesis is to research ANN model and use it to personal credit assessment, the focus is to improve the tradition personal credit scoring model by using BP ANN. There are two main things in this thesis:One is creating four BP ANN models which adopt different algorithms. The other is creating the weighted average BP ANN model in order to improve the generalized ability of BP ANN model.The research of BP ANN is thorough and its algorithms are numerous. For different algorithms, we can’t comment this one is better than the other one simply, but we can find the most suitable method for every specific question. In this thesis, I use four different algorithms. The result displays the models of trainbfg and traincgp are more suitable than the models of traingdx and traincgp to the problem in this thesis. Comparing all aspects, the model of trainlm is most suitable in four models.In order to improve the generalized ability of BP ANN model, I create the weighted average BP ANN model. This method is mainly based on two principles:one is we must fully tap the information contained in the samples and reduce the impact of the noise data; the other is BP ANN has the trend of forgetting the old samples in the process of learning new samples. The rule of the method is as follows: firstly we need create n data sets by randomly sample from the test data set, then create BP ANN models based on every data set and apply them to the test data set, last we use the weighted average result as the final result. The result show that the weighted average BP ANN model can improve the generalized ability of BP ANN model.