Study On Logistic-RBF Combination Model Of Personal Credit Scoring

Along with the Chinese economy fast development, the personal credit consumption also gradually becomes one of the main individual consumption forms. The fast growth of personal credit consumption requests the bank to have perfect credit risk management system. However, lacking the scientific personal credit scoring system is one of most serious problems existed in the present Chinese commercial bank credit risk management. The main reason to explain why many personal credit scoring models can not be popularized is that the stability of the models is not so good while the accuracy is improved, especially the ability to deal with the second rate of misjudge is week. This dissertation will solve this problem using the combination model thought.Based on the domestic and foreign scholars'research, analyze the principle and modeling thoughts of single models and combination forecast models, and select ten representative personal credit scoring index. At the same time, it extracts five groups of samples from our Chinese commercial bank data, which are used in training and testing. Using the high stability advantage of statistical method of Logistic regression and high accuracy advantage of non- statistical method of Radial Basis Function neural network, this dissertation establish fixed weight combination forescasting model and variable weight combination forescasting model in order to synthesize the different advantages of the single models and improve the accuracy and statibly of combination models.Meanwhile it analyze the choice and applicability of single models and combination models, and then it compare the single models and combination models and find that: the stability of Logistic regression is 0.0071 which is higher than RBF neural network 0.015984, but the accuracy of RBF neural network is 94.77% which is higher than Logistic regression 92.47%; The accuracy, stability and the ability of dealing with the second evaluation problem of combination models are higher than the two single models; the accuracy and stability of variable weight combination forescasting model are 0.0042 and 95.23% which are higher than fixed weight combination forescasting model. Therefore, in personal credit...