Research On Prediction Of Overdue Loan Risk Based On P2P User Data

P2P(Peer to Peer lending),is developing rapidly in China nowadays.However,the problem of high default rate of borrowers is becoming more and more serious,which has caused great losses to the platform and creditors.In this case,how to establish a perfect risk prediction system is the decisive factor for the smooth development of P2 P platform.Due to the late start of domestic P2 P industry,many companies are relatively weak in this respect.Therefore,it is an urgent problem for the development of P2 P to design a model with good forecasting ability and strong generalization ability to construct a risk prediction system to evaluate the credit of borrowers,and then to predict the overdue loan and reduce the default rate.At present,existing platforms,such as PPDai,LUp2 p,are strengthening their own risk prediction model.In order to solve the above problems,this paper constructs a prediction model for the overdue risk of P2 P borrowers.The main work of this paper includes:(1)The construction method of risk prediction model based on fusion is proposed and the basis theory is elaborated.(2)On the basis of the user data set of the P2 P lending platform,the problem of data optimization and class imbalance is solved by data preprocessing and SMOTE algorithm improvement.The processed data are extracted,constructed and selected,and the optimal feature subset is constructed,which lays a foundation for the follow-up study of risk prediction model.(3)The commonly used risk prediction models are studied and compared,and the model is improved and adjusted according to the characteristics of the feature subset of P2 P.The experimental results show that the risk prediction model composed of XGBoost,random forest and support vector machine has good prediction effect,which provides the basis for the selection of the algorithm for the next step of the fusion risk prediction model.(4)After comparing the methods of model fusion,stacking ensemble learning framework is selected to fuse XGBoost,random forest and support vector machine.The prediction accuracy and generalization ability of risk prediction model are improved by empirical fusion.It fully proves that the risk prediction model proposed in this paper plays an important role in the risk control of P2 P.In summary,this paper analyses the key points and difficulties of P2 P risk forecasting model,puts forward a method of constructing risk forecasting model based on fusion model,constructs the optimal feature subset of the model,compares,selects and improves three algorithms,and applies them to the construction of model fusion risk forecasting model,and proves the superiority of risk forecasting model based on fusion through experiments.The practical scenario of the model is also discussed.