The Bayesian Additive Classification Tree Applied To Evaluate P2P Borrowers' Credit Risk

Online Peer-to-Peer lending,known as P2 P lending,started in Britain,with the development of Internet Technologies,it rapidly develops at home and abroad in recent years,and has all kinds of operation patterns.P2 P lending belongs to micro-finance,praised by people for its low investment threshold.However,it is also widely attention for bad events such as platforms run,borrowers default and so on,emerging in endlessly.The research object of this paper is China's P2 P lending.We use two commonly used machine learning methods and a new method of Bayesian to evaluate P2 P lending borrowers' credit risk,these three methods are : Support Vector Machine(SVM),Random Forests(RF)and the Bayesian Additive Classification Tree(BACT),respectively.Firstly,we split raw data into training data and testing data,then divide training data into three same sections and call it raw training set.Considering our data has class imbalance problem,so we use SMOTE algorithm to process raw training set and call it SMOTE training set.Secondly,we use raw training set and SMOTE training set to pass 3-fold cross-validation and select optimal parameters corresponds to models simultaneously.Thirdly,we combine training set,selected parameters with models to predict testing data and compare classification performance between different models under different training set.Finally,we find that when using SMOTE training set to train model,BACT and RF are more able to identify defaulting borrowers,but their AUC value don't change significantly and the AUC value of SVM is significantly larger than before;The accuracy and AUC value of BACT are bigger than RF and SVM,and its two misclassification rates are smaller than RF and SVM;The ROC curve corresponding to RF and BACT are obviously located in the upper left of SVM's corresponding ROC curve,the ROC curve corresponding to RF almost completely coincide with BACT's corresponding ROC curve,all above show that BACT has good performance in evaluating borrowers' credit risk under SMOTE training set.