P2P Borrowing Risk Prediction Based On Machine Learning

P2P online lending is an innovation of the financial industry using the network to make loans to traditional Banks.It connects borrowers and investors through the Internet.In recent years,P2 P online lending has developed rapidly,but the emerging problems are also obvious,that is,the default situation of borrowers is widespread.This paper makes a risk prediction model for the information of borrowers provided by P2 P platforms,and compares and analyzes several models.First carried out a statistical analysis on the data provided by the platform,make the feature selection and feature of the restructuring,and building Decision Tree model?Random Forests model,Adaboost model,found that while the overall prediction error is low,but the small samples of the prediction error is high.Then,the unbalance data is based on the random undersampling of clustering,and then the Decision Tree model,Random Forests model and Adaboost model are constructed respectively for the obtained equilibrium data.Finally,in order to pay more attention to the prediction of default users,this paper will improve the weight of the default users and retrain the classifier under the changed sample distribution.The comparison of generalization error and confusion matrix of the models obtained in the above three cases are presented.The results show that after balancing the data and changing the weight of the sample,the overall error is improved a bit,but the classification of small samples is more accurate.