Research On Behavior Of Lenders And Borrows In P2P Lending Platform

P2P(Peer to Peer lending))is an Internet service website used by individuals to borrow money from individuals,namely,private lending and network technology.P2 P network platform provides a high yield financial channel for financial managers,and provides a convenient financing way for individuals and small enterprises.The data of the second half year of 2016 is considered as the original data to analyze the intermediary functions of P2 P lending platforms.Upon analyzing the lending behavior of investors,we can identify different of them,so as to increase the lending funds of P2 P lending platforms.Then the loan behavior of the borrower is analyzed to effectively identify the default behavior of the borrower.Strengthen the platform's behavior management to the borrower,in order to reduce the probability of loan default occurrence.By introducing data mining algorithm and exploring the behaviors of lenders by association rules analysis,we found that,to determine whether the new lending user "valuable users" and"temporary users",the most important factor is the second and third months from the first time-point of loan.Meanwhile concerning to the borrower,we have the conclusions that:(1)Upon using correlation analysis and random forest,13 important variables are selected from 17 and the most distinguished three factors are the times of successful borrowed,the magic mirror grades and the interest rate of loan.(2)By the comparison of the variables selected,combining with the different data mining algorithms(decision tree and support vector machine(SVM),Bayes algorithm and regression tree,logistic regression,Bagging,Boosting),it shows that the Boosting method based on decision tree has the highest prediction accuracy.(3)To deal with the imbalance of the data,a mixture model of "clustering,data processing and classification of the hybrid model''is constructed,namely the PAM clustering algorithm + level for each class data processing(weight,less sampling,sampling,SMOTE)+ Gradient Boosting based on decision tree with 10-fold cross validation.It is illustrated that the hybrid model of "clustering + data processing + classification"proposed has effectively promotion to the prediction of borrower's default.