Research On Credit Evaluation Of P2P Borrowers Based On Random Forest

It has been ten years since P2 P lending appeared in China,and it has become an important financing channel for individuals and minor enterprises.Before the end of 2016,the total number of China's p2 p lending platforms has reached 5877,while only 2448 of them are in normal operation.Meanwhile,the volume of P2 P lending has reached 206.3872 billion yuan and its overall loan balance has reached 816.224 billion yuan.Although the number of normal operation platforms has been reduced,compared with 2015,last year's turnover and loan balance had increased by 110% and 100.99%,respectively,which indicates that P2 P lending is still very important in financing.With the development of regulatory overhaul in China,the external environment of P2 P lending has been improved gradually.However,the credit problem of borrowers in P2 P lending platform is still severe,which affects the sustainable development of the industry.Furthermore,most of the domestic P2 P network lending platforms lack effective credit assessment of borrowers,which strengthen borrowers' default risk in turn.Therefore,this paper constructed the credit assessment model based on random forests and data from the pterosaur loan platform to evaluate the borrower's credit risk.In the study,the main concepts and background of P2 P network lending are explained at in the beginning.Then,the random forests model,vector machine model and Logistic regression model are introduced.As a result,the forests model is selected by comparison.Finally,with the data captured from the pterosaurs platform,this study constructed credit assessment system,and testify its efficiency through comparative analysis.In the process of constructing the random forests model,19 indicators of the credit risk for borrowers are selected firstly based on the research in the P2 P lending.Then,the selection process is carried out based on the random forests,which finally determines the risk index system with 8 indicators.Furthermore,the paper use the R language to build random forests model and support vector machine,and thus compare the classification effect.The results of empirical analysis show that the selection criteria based on random forests feature can improve the classification performance of the model,and the random forests model is superior to the support vector machine in the credit risk assessment of P2 P lending.The results also shows that the borrower's historical information in the platform and the information of the loan object are important indicators to control the borrower's credit risk.