Application Of Generalized Partial Linear Variable Coefficient Model In Credit Evaluation

With the economic growth,the business of personal loan is increasing day by day,and the gradual increase of overdue behavior needs to be solved urgently.Credit evaluation model,as one of the important tools for banks and other lending institutions to control credit risk,is an important research problem.Statistical methods have been widely used in credit evaluation.This thesis mainly applies the generalized partial linear variable coefficient model.The statistics models are established,and the factors affecting customer default are analyzed.Firstly,the B-spline basis function is used to approximate the non-parametric function coefficients of the generalized partial linear vary coefficient model.Lasso method and Group Lasso method are used to select variables and estimate parameters simultaneously,and coordinate descent method is used to solve the model parameters.Group Lasso and Lasso methods are used to do numerical simulation.Secondly,pie charts and histograms are used to depict variables intuitively.Dummy variables and data standardization are introduced to preprocess the data.Considering the unbalance of the data itself,Random Oversampling method is used to supplement the data.The generalized partial linear variable coefficient model and Logistic regression model are used to model,and the group Lasso and Lasso methods are used to select variables and estimate parameters simultaneously,and carry out compare analysis.Combined with the non-linear form of function coefficient estimation and the comparison results of ROC curve,explanatory ability and flexibility of generalized partial linear variable coefficient model are analyzed.Compare with the Logistic regression model in the ability of accurate default prediction,our model has superiority.Finally,according to the results of statistical analysis,some suggestions for the credit evaluation are offered.