Comparative Analysis Of Neural Network In The Credit Risk Assessment Of SME

In recent years, the rapid development of small and medium enterprises (SME) has become an important force in promoting China’s economic development. But the financing difficulties are the bottleneck during the development. Efforts from many aspects should be done to solve the difficulty of SME financing. To improve the accuracy of credit risk assessment to the SME; to make excellent small and medium-sized enterprises get loans in time and to reduce the credit risk faced by the commercial banks are very important aspects of solving the problem of financing SMEs.Credit risk is the probability of the dealing party cannot perform the obligations, including the impact on income and investment funds of financial institutions caused by the reducing of counterparty credit rating, leading to financial difficulties of the bank. The credit risk assessment needs the modeling techniques of uncertain events state of the art modeling. Accordingly to which, credit risk is the main threat faced by the financial institutions. It is also one of the elements of modeling financial institutions in financial difficulties.As one of the most common methods of artificial intelligence, neural network model has been widely applied in credit risk assessment, and has obtain better results. However, the application of neural network model is rarely used in credit risk assessment of the unlisted SMEs in China, and there is no systematic comparative analysis. Because credit risk assessment involves many factors, the traditional neural network model has been difficult to meet the needs in the field of in-depth exploration and research, and the complexity of the credit risk assessment is also an urgent need for multidisciplinary cutting-edge technology and theoretical cross practice.In response to these problems, the paper briefly reviews the application status of the neural network models in credit risk assessment, and gives a brief description of several common neural network models. This paper first compares the effect of several common neural network models used on the small and medium-sized enterprises in our data set, and determines the optimal parameters of each model. Then the paper compares the classification accuracy and applicability of the models.Based on of unlisted SME data sets, the paper presents the experimental comparison of classification accuracy and applicability of several common neural network models, and raises the corresponding recommendations to the concrete application of credit risk assessment model. This aims to make some contribution to solve the problem of financing SMEs.