Research On Credit Score Model For Credit Card Customer

With the development of credit business and the change of consumption behavior and consumption idea,credit card business has developed rapidly.But the credit card business has credit risk,interest rate risk and other risks,and in China credit system started too late and far behind the developed countries.Setting up credit score model and controlling the credit risk is not only the need of theoretical research to explore,also is the need of social economic practice.Based on the principle of combining qualitative analysis with quantitative analysis,in this paper,we introduce the current situation of the development of credit card,the necessity of raising credit score model,and the credit score commonly used models and practical application situation in developed countries.Based on the date of 1000 credit card customers in June 2013 to December 2013,and the date include a month overdue status,monthly payments and amounts as well as other available information,including the default condition of each customer in all of 2014.On the basis of the sample data,we use the statistical and data mining technology,using Logistic regression model,the linear discriminant model,Decision tree,support vector machine(SVM),random forests and Adaboost lifting method.Followed by data preprocessing,variable selection,new variables adding,correlation analysis and processing the redundant,several kinds of credit rating model is established,and we compare the evaluation of these models from classification accuracy,the second type of error rate,the stability of the model and the ROC curve,AUC value.The results show that the model precision of random forests and Adaboost method is better,but poor stability,and while the accuracy of Logistic regression model is slightly weaker than the two models,it has good stability.