Research On Credit Risk Of P2P Network Loan Platform Borrowers Based On Tree Intergration Model

P2P(Peer to Peer)means personal-to-personal.P2 P online lending is a subindustry of connected finance.It is a business model in which individuals and individuals directly realize mutual lending transactions through online lending platforms.In recent years,the development of P2 P online lending industry is very hot at home and abroad,and various types of online lending platforms have also emerged.Especially in China,according to data from WangDaiZhiJia,by the end of 2018,more than 6,000 P2 P online lending platforms have been accumulated.However,some bad phenomena such as runaway and suspension of business have also broken out on P2 P online lending platforms,among which a large number of defaults by platform borrowers will seriously affect the stable development of P2 P online lending industry.Therefore,this paper focuses on the credit risk of online loan platform borrowers,and establishes a rapid and effective pre-loan credit risk assessment model for online loan platform borrowers,so as to promote the healthy development of P2 P online loan industry in the future.This paper first introduces the integration model of three kinds of trees to quickly and effectively evaluate the credit risk of borrowers on the online loan platform,namely Gradient Boosting Decision Tree(GBDT)model?the Random Forest(RF)model and Extreme Gradient Boosting(XGBOOST)model.Then,by obtaining the borrower's loan data of the renrendai platform,the pre-processed data are randomly divided into the original training set and the original test set.Due to the imbalance of the borrower's online loan data,the BLSMOTE training set is obtained after the Borderline-SMOTE algorithm is used to balance the original training set.Before the modeling and prediction of each model,the 5-fold cross-validation method was used to optimize the model parameters and select the optimal value.The corresponding classification evaluation index and running time were obtained from the training prediction of each model under the original training set and BLSMOTE training set,respectively.By comparing the classification evaluation index and running time of each model,the conclusion is drawn: under BLSMOTE training set,the classification performance of XGBOOST model and GBDT model is improved,while that of RF model is slightly decreased.Under the original training set and BLSMOTE training set,the classification performance of XGBOOST is better than that of GBDT model and RF model,in which the ACC value and AUC value in the classification evaluation index are larger than the latter two,and the corresponding ROC curve is closer to the upper left point(0,1)than the latter two.The XGBOOST model predicts that both FP and FPR(the first type of misclassification rate,which is the proportion of people with poor repayment mistakenly divided into those with better repayment)are smaller than the GBDT model and RF model.In terms of running time,XGBOOST model consumes the shortest time.Therefore,this paper believes that XGBOOST model is a fast and effective credit risk assessment model for online loan platform preloan audit borrowers,which can provide reference value for platform managers to effectively evaluate and manage.