Application Research Of Personal Credit Risk Assessment Based On Pre-loan Behavior Data

With the development of Internet technology,the P2 P line Internet loan platform has developed rapidly.However,in recent years,many problems have arisen in the P2 P industry,such as platform information transactional opacity,imperfect guarantee mechanism,improper pre-loan approval measures,and inadequate post-loan tracking.The core reason lies in the imperfect personal credit evaluation in the pre-loan approval process.Therefore,this paper uses data mining methods to mine the partial buried data in 2018 of a loan company's APP,i.e.user's pre-loan behavior data,is extracted and derived.The data are processed by means of equalization box and binary tree sub-box.The importance of features is sorted and filtered according to WOE and IV values.Then the logistic regression model and the xgboost model based on the binary tree and the equal bin are established,from which personal score card is established based on binary tree logical regression model.The results show that: Firstly,in the two binning methods,the logistic regression and xgboost model based on the binary tree binning method have the greatest distinguishing ability between positive and negative samples;secondly,compare the xgboost model with the logistic regression model,no matter which binning The performance of the logistic regression model is slightly lower than that of the xgboost model.However,by constructing a behavioral score card,the logistic regression model can more intuitively reflect and quantify the risk level of the customer and improve the explanability of the risk.Finally,among the many indicators of degree,equipment safety,and special events,the characteristics that have an important impact on personal credit risk assessment are special events,including the user's touch screen sliding behavior on each page of the APP.The research in this paper shows that,under the condition that China's credit information system is imperfect,it is helpful to evaluate the pre-lending risk of credit applicants and classify them by collecting credit applicants' pre-lending APP behavior data.