The Research Of Personal Credit Evaluation Based On Logistic Regression And Probit Regression

Recently,due to the change of personal consumption concept,China's consumer credit has entered a period of rapid development.The data show that in 2017,the credit balance of the whole industry reached 104.1768 billion yuan,an increase of 62.2% over the previous year.it is showing an explosive growth trend.With the development of Internet information technology,P2 P network loan platform has become the representative of Internet financial investment.This mode enables customers to apply for loans quickly through simple auditing,which brings great convenience to people.Although the P2 P network loan can solve the financing problem of micro-lending customers,the default rate of borrowers is rising.Investors often fail to accurately judge the credit level of borrowers,which damages the interests of investors.Therefore,the study of credit evaluation by borrowers has become a key issue.Firstly,this paper introduces the development background of credit evaluation and the current situation of P2 P network credit,then introduces the commonly used methods of building credit risk assessment model,including mathematical statistics and artificial intelligence methods,with emphasis on the theoretical basis of Logistic regression and Probit regression.Taking Lending Club Company in the United States as an example,this paper selects some borrowers' information data of the company to analyze the credit risk.Personal credit risk assessment is studied.In the construction of Logistic regression and Probit regression models,data preprocessing is carried out.Firstly,missing values are processed by deletion and mean interpolation,then data imbalance processing and data sampling are carried out.Finally,feature variables are roughly screened and carefully screened.We introduce the concepts of WOE and IV values,and select 14 values whose IV value is greater than 0.02 and may be related to credit risk assessment.Then we do descriptive statistical analysis and correlation analysis,we delete two variables in the correlation analysis,and exclude the multiple collinearity of the model.Then,a personal credit risk assessment model is constructed by Logistic regression and Probit regression,and the predictive ability of the model is tested by discriminant matrix,ROC curve and AUC value.The results show that Probit regression,like Logistic regression,has good predictive ability.Because the model results are difficult to explain,the application of Probit regression is not popular,but Logistic regression can not completely replace Probit regression.At the same time,the Logistic regression equation and Probit regression equation are composed of 13 indicators,such as loan interest rate,loan purpose,working life,validation status,loan status,annual income and so on.The model is effective.Lending Club company,as the originator of the network credit circle,can provide reference value for the credit risk research of P2 P network credit in our country,and even for the healthy development of the whole Internet finance.