Credit Risk Assessment Model Based On Feature Generation And Historical Records

In recent years,P2P(peer to peer)lending has become one of the core formats of Internet finance.However,the default behavior of borrowers has caused the platform to suffer the loss of funds and is not conducive to the orderly development of the market.Therefore,it is imperative to conduct a scientific and reasonable default forecast of the borrowing behavior.The existing risk assessment models often require experts to perform complex feature engineering and make use of the borrower's current loan information for risk prediction.The method is time-consuming and difficult to reveal the borrower's default intention.Therefore,we need a more efficient and accurate risk assessment model.In this paper,by introducing the feature generation and considering the borrower's historical records synthetically,we propose a Feature generation&Historical records based Risk assessment Model(FHRM)to solve the problem of default forecast.First,we use GBDT algorithm to generate the features of borrowers,bidding information,automatically find effective combinations of features and features to make up for the lack of artificial experience and shorten the experimental period of the model.Then,according to the borrowers' historical loan records,we learn about their borrowing behavior through long short-term memory network and find a high-risk sequential pattern.Finally,we train the FHRM model to obtain the borrower's risk assessment result.The experimental results on real data set Prosper show that credit risk assessment model based on feature generation and historical records can improve the accuracy of forecasting results by using feature generation technology and long short-term memory network,which can provide a scientifically effective credit risk assessment method for P2P online loan market and provide investors with an important basis for investment decisions.