The Research On Customer Default Forecast Model Of The Commercial Bank

Nowadays,with the rapid development of economic,non-performing loans of commercial Banks are rising in our country,which directly affects the social and economic activities.So people must control credit risk.Cycle of traditional artificial credit audit is long and inefficient,which is very difficult to guarantee accuracy.In recent years,with the development of data mining and machine learning,through detailed interpretation about the domestic and foreign scholars for the commercial bank credit evaluation,we found that most of the credit evaluation model has great subjectivity in terms of variable selection and indexes are one-sided,lack of attention to default customer identification.Therefore,it is imperative to establish credit evaluation model of the high stability,high accuracy.Based on the existing research,using data mining technology,we consider the variable selection and default model building,build the relatively stable and highly recognized the default customer personal credit evaluation model of the rate.We use 2016 CDA competition data for empirical analysis,introducing the Earth Mover's Distance(EMD)method for the choice of variables and on the basis of the decision tree model considering two kinds of error causing loss to the commercial bank,we build a decision tree model with cost-matrix.Based on F2-score index,we compare logistic regression,random forests,support vector machine and neural network model to test the effectiveness of our model in identifying default clients.In terms of variable selection,compared with T test and IV value,the EMD has the highest accuracy and F2-score.Compared with the four traditional default model,our model also has the highest F2-score.In our model,9%of the loan defaulters were wrong,and random forest model has 23%of the loan defaulters were wrong.Our model show on the basis of guarantee the accuracy,our model reduce the second category mistake rate(default customer is identified as normal),which has significant reference and application value for commercial Banks' construction of personal credit evaluation system.