Study Of Credit Evaluation Model Based On Neural Networks

The credit risk is harmful to finance market. It is the loss possibility which is caused by business opponent or debt publisher falls back or change credit character in finance business. The credit risk influences all kinds of activities in modern economy life immediately, macro decision-making and economic development of our country, global stable economic development. So it is a new issue of finance institution that establishing a correct and effective forecast model which evaluates the enterprise or individual credit and confirm a kind of effective settle scheme and reduce bad loan rate.The paper first analyzes the disadvantage of the credit model and all kinds of methods which are now used in credit evaluation, and then combine the specialty of credit evaluation, uses artificial neural networks technology, establishes enterprise credit model and individual credit model which are both based on neural networks. The model provide forceful support for how to establish a scientific, objective , correct and feasible credit evaluation model.This paper analyzes the enterprise and individual credit evaluation phenomenon, it clarifies that these models and methods are not enough to reflect the non-linear relations between many factors which influence enterprise or individual credit. Then by analyzing the neural networks technology of AI, it finds that neural networks is a natural non-linear model procedure and finds rules from a lot of complex data which are efficacious specialty for credit evaluation. So it is feasible to establish credit model with neural networks technology.By setting up index system and getting index data from bank, the paper simulative experiments on the credit evaluation model based on BP networks. In the enterprise credit evaluation part, different methods arc used to process the flexible and hard index data. In the individual credit evaluation part, it quantifies all kinds of scattered data, and to improve BP algorithm it uses local self-adaptive study rate algorithm. The result indicates that the real output is almost equal to the expected output, and the model has a high correctness and practicability. So it has a high value of study and generalize application.