Research On P2P Lending Risk Model Based On LightGBM

With the rapid development of Internet finance,the concept of P2 P has also swept across the country in recent years.However,the extremely high bad debt rate has brought huge losses to the P2 P platform operators and investors.Therefore,in order to improve the healthy development of the P2 P industry,it is necessary to establish an efficient loan default risk identification model.This paper aims to build a model that can effectively identify the default risk of P2 P lending based on the machine learning method.It can help to reduce the default risk of P2 P companies,guarantee the profits of companies and investors,and provide suggestions to improve the development of P2 P industry.At present,most research methods of P2 P lending default model are based on logistic regression or decision tree.However,for high-dimensional and massive date,the prediction accuracy of these methods are not high.The LightGBM model helps to improve the prediction accuracy and gains profits for the P2 P platform.This paper collects the data of Renrendai,performs descriptive analysis,cleans data and filters features.After that,a variety of machine learning models are established for the processed data sets,and the prediction performances of each model are evaluated and compared by the evaluation indicators such as accuracy.Finally,we select the LightGBM model as the base model,and use Linear Blending to integrate the model to establish the final model.The final results of this paper show that the LightGBM algorithm is superior to traditional models such as logistic regression or SVM for high-dimensional and large-scale data.After the model integration by Linear Blending,the prediction performances can be further improved,which is better than the traditional model.The integrated model can effectively identify the default risk of users so it could be applied to the P2 P platforms for risk control.