Research On Credit Risk Prediction Of P2P Borrowers Based On Data Mining Technology

With the development of economy and the progress of science and technology,China's Internet finance has achieved unprecedented development,and as one of its important models,P2 P online lending has also emerged.Compared with traditional lending,P2 P lending transactions are no longer limited by time and place,which greatly improves the convenience of lending transactions and the utilization rate of social funds.However,P2 P lending,which is still financial in nature,instill fresh vitality into China's economy at the same time,many risks inevitably arise,among which the most severe and most concerned by academia and society is the borrower's credit risk.Therefore,how to effectively predict the credit risk of P2 P online loan borrowers has important theoretical and practical significance.Based on data mining technology and taking Prosper platform borrowers' open data set as an example,this paper studies the credit risk of P2 P online lenders.Firstly,on the basis of relevant literature research,this paper gives a brief overview of P2 P lending,P2 P lending in China and its development history and current situation.Then it discusses the theory of data mining and the theory of P2 P network credit risk,and constructs the data mining model based on the random forest model.Then,the data set is analyzed and preprocessed.Then,the random forest model and other comparison models,such as Logistic regression model,support vector machine model and naive Bayesian model,were used to train and predict the borrower data before and after the platform adjustment.At last,the author evaluates and summarizes the effect of each model on the comprehensive prediction of borrower's credit risk.The specific research conclusions of this paper are as follows:First of all,in terms of the credit risk prediction and control method for P2 P network loan borrowers,all data mining models can effectively predict the credit risk of borrowers to different degrees.Among them,the integrated learning model--random forest overall prediction has the best effect and the most stable comprehensive performance,which can be first used for the credit risk prediction of P2 P network loan borrowers.Therefore,the construction and application of the borrower's credit risk prediction model can effectively promote the construction of credit risk control in P2 P lending industry.Secondly,the random forest feature selection method is used for data dimensionreduction.Compared with the principal component analysis method for data dimension reduction in most studies,the computational efficiency is higher and the result of feature selection is more conducive to the later model establishment.At the same time,as a feature selection method,random forest has a good effect.It can automatically select features of high importance and draw the results into a visual chart.This method is conducive to the selection of input variables in the later model,so as to improve the model's ability to predict borrowers' credit risks,and help the P2 P online lending industry to better adapt to the development needs of current stable finance.Finally,the "credit rating" variable can greatly affect model performance.The accuracy of each model after July 1,2009 is much higher than that before July 1,which indicates that improving the variable of "credit rating" can effectively improve the prediction accuracy of the model.This variable adjustment has certain effect.Therefore,P2 P lending platforms can employ professionals to strengthen the collection and prediction of "credit rating" information,optimize the borrower's credit evaluation indicators,so as to achieve scientific prediction and prior control of the borrower's credit risk,and thus promote the platform's healthy and long-term development.