Credit Scoring Model Using Fuzzy Neural Network System

Credit scoring is one of the important methods to predict financial trouble for listed companies. It is benefit for the bank to analyze risk if we can precisely quantify the value of credit risk. This paper does research on two and three pattern recognizant of the listed company in Shanghai and Shenzhen.In the two patterns recognizant, we choose the financial data of 106 listed companies at 2000. 53 of these companies are in financial distress, while the other 53 companies are in normal situation. The data sample is divided into two parts: 63 training data and 43 testing data. After eliminating some abnormal data, the sample data are input into fuzzy neural network, BP network and Elman network respectively. The result shows that the classification and prediction ability of fuzzy neural network and Elman network are better. Both of them misjudge one listed company. Elman network classify a company into 'Bad' credit while it is in good situation, and Elman neural network classify company into 'Good' credit while it is in financial distress. As for Bp network, it misjudges four companies. The training error and accuracy degree of fuzzy neural network is 0.072 and 97.3% respectively. The empirical analysis demonstrates that fuzzy neural network performs better than traditional Bp network for credit scoring.Further, in the three patterns recognizant, our sample is the financial data of 96 listed companies at 2000. We divided them into three types: good, medium, bad. The classification result shows that the accuracy of ANFIS reached 74.2%. By increasing the number of membership functions, ANFIS performs better so that the accuracy can reach 87.01%. While comparing the performance of ANFIS with discriminated analysis, we find that ANFIS is more accurate than discriminated analysis, however, discriminated analysis perform better than ANFIS to identify bad credit applications.To conclude, this paper has four parts: in the preface, it introduces the current advance and problem of study on credit scoring. Also the development of credit scoring model with fuzzy neural network is described. The second part is about the structure of fuzzy neural network. We describe the detail step by step, including fuzzy inference system and the 5 layer structure of ANFIS. When it comes to the third part, a detailed introduction is given to Hybrid Learning Algorithm, which is used to training the network. Hybrid Learning Algorithm is divided to two procedures: the forward pass with Least Squares Estimate and the backward pass with Gradient Descent. The last part is the application example. We explain the choice of index and classification result. Then it comes to this paper's conclusion.