The Research On Borrowers' Credit Risk Assessment In Peer-to-peer Lending Based On Random Forest

Peer-to-peer(P2P)lending integrating the Internet technology with the financial technology is entering the stage of the development of compliance from the development of expansion,which has become one of the important channels to solve the problem of short-term microfinance liquidity in the loan market.In the bilateral market where take P2 P lending platform as the ligament,information symmetry has not yet been realized between the borrower s,the platform and the lenders.In addition,most borrowers served by P2 P lending platform have no records in the CCRC consumer credit report.Therefore,how to identify the borrowers with lower cost,more efficiency and more precision is the core problem that scholars and industry insiders have been working on for several years.The thesis uses the frontier random forest(RF)algorithm,constructing the credit risk assessment model of borrower under P2 P lending environment.What's more,multi-classification is made on the borrowers' credit risk rating,the applicability of this machine learning model to P2 P lending market is explored,and the new research direction is provided for the credit risk control of the borrowers by the P2 P lending platform.The thesis takes the borrower's credit ratings as the target variable to build the borrower' credit risk assessment model by RF algorithm in the P2 P lending,and uses the 17,917 observation samples from debit and credit transactions data in renrendai.com.In the process of feature engineering,the index system is constructed by using the important()function,varSelRF package and feature selection method in RF model.In modeling section,as to the imbalanced distribution characteristics in the original dataset,we apply Synthetic Minority Oversampling Technique(SMOTE)for increasing the number of cases in original training dataset in a balanced way,after optimizing the parameters,the optimal parameter combination is determined successfully.That is,the optimal number of random partition variables is 6 and the optimal decision tree number is 1000.Accordingly,the RF model for borrowers' s credit risk assessment in P2 P lending is obtained based on the optimal parameter combination.The thesis results show that the random forest algorithm is suitable for the borrower's credit risk assessment in P2 P lending and shows a good classification performance.Among the 35 candidate variables,the FieldCert and the InterestRate are especially important,which is instructive to the lending decision of the peer-to-peer lending platform and the investors.Finally,the evaluation results of the RF model based on the optimal parameter combination show that the borrowers' credit rating obtained by the credit risk assessment model is more rigorous than that the renrendai.com built.