Classification Prediction Of P2P Network Loan Default Based On Hybrid CatBoost

As a representative of internet finance,P2 P online lending develops rapidly.At the same time,many risks are exposed,the most important of which is credit risk(borrower default risk).Therefore,it is of practical significance to classify and predict the borrower's default and study the factors affecting the borrower's default.This paper selects renrendai,a representative P2 P online lending platform that is launched earlier in China,as the research object,and selects its effective historical transaction data in 2016 and 2017 to build binary classification prediction models for empirical research.First,this paper constructs 19 features based on five aspects.Then,random forest is applied for feature selection,and the top 15 important features are selected to participate in modeling according to the average score of every feature;Secondly,different imbalanced data set processing methods(undersampling,oversampling,undersampling and oversampling combination)and machine learning algorithms(SVM,RF,AdaBoost,XGBoost,LightGBM,CatBoost)are combined to establish the network loan default classification prediction model;Finally,AUC and other indicators to measure the performance of the classification model are compared and analyzed,the results show that the SMOTETomek-CatBoost model of hybrid CatBoost models has better classification prediction performance than other models(SVM and hybrid SVM,RF and hybrid RF,AdaBoost and hybrid AdaBoost,XGBoost and hybrid XGBoost,LightGBM and hybrid LightGBM).By analyzing the feature importance score results of hybrid XGBoost,hybrid LightGBM and hybrid CatBoost models,seven main factors affecting borrower default are obtained.