| In recent years,with the rapid development of the Internet,the Peer to Peer Lending has entered the public's vision and attracted many borrowers and investors with its own characteristics of low threshold,high profit and convenient operation.Compared with foreign P2 P lending,the P2 P lending in China can follow the global trend,but the concept of risk control is insufficient.As the number of problem platforms is increasing,and more borrowers and investors are involved.Therefore,how to improve the risk monitoring of P2 P online lending platform,further enhance the credit evaluation of P2 P online lending platform to borrowers,and reduce the investment risk of investors,is a very important issue for the future development of P2 P lending platform.In this paper,we use the traditional classification methods of logistic regression model and machine learning,including random forest,neural network and support vector machine,according to the foreign lending club website from2016 to 2017 more than 70,000 new customer loan data for analysis.The results show that the four models have high classification accuracy,and the classification accuracy is more than 80 %.However,due to the imbalance of the data,the classification accuracy of the four models for bad customers is not ideal,all of them are about 2 %.This is the objective reality that many researchers are facing.In order to improve the classification accuracy of bad customers,this paper optimizes the imbalanced data set by SMOTE,and then fits and the data optimized by SMOTE again through the above model.The results show that both the random forest model and the support vector machine model have good prediction effect,and the classification accuracy of bad customers is improved to more than 85 %,which is obviously superior to logistic regression and neural network.The results show that SMOTE optimization method,random forest and support vector machine model are effective methods to deal with complex unbalanced dataset. |
PDF Full Text Request