Research On P2P Borrower Default Prediction Based On RKM-GP-Boosting

P2P network lending is an unsecured credit loan model based on the Internet platform.It can collect small amounts of scattered funds from the society and lend them to borrowers or enterprises with needs,so as to achieve rational redistribution of economic resources.However,the rapid development of P2P network lending also brings many unsettled factors.The information asymmetry between borrowers and lenders may bring serious consequences of loan default,resulting in frequent bad debts and running.This not only gives huge losses to borrowers and P2P lending platforms,but also casts a shadow over the development of the P2P industry and even the Internet finance industry.Firstly,this paper constructs a P2P borrower default indicator system from five dimensions by referring to domestic and foreign literatures and combining relevant information of the platform,which are the borrower's economic situation,credit evaluation status,historical borrowing status,basic personal information and borrowing text attributes.Then,using the feature selection method of filtering,wrapping and embedded to screen the index characteristics,the two dimensions of historical borrowing status and credit evaluation are significantly affected by the borrower's default forecast.Secondly,for the P2P borrower default prediction problem,the data set imbalance and the boundary sample have strong influence characteristics.An undersampling method based on neighborhood rough set Kmeans clustering is proposed and combined with the Boosting class lifting algorithm.The framework of RKM-GP-Boosting algorithm for P2P borrower default prediction is proposed.Finally,this paper uses the Lending Club platform to obtain model performance verification from 2,132,200 loan data from June 2007 to the end of September 2018.The evaluation results show that the classification performance and stability of RKM-GPBoosting algorithm are significantly better than that of Smokee,Easy Ensemble and other unbalanced classification algorithms.This model framework can provide decision support for lender investment and platform supervision in P2P industry.