Research On Early Risk Warning Of P2P Network,Loan Platform Based On Classification Algorithms

In recent years,the rapid rise of emerging financial industry represented by P2P network lending has promoted financial innovation.However,due to industry system and information asymmetry,platform security issues have frequently occurred.Overdue and illegal fund-raising fraud are serious.P2P online lending platform become the hardest hit by financial fraud.Due to the wide range of Internet finance and strong correlation,once a large-scale default event occurs on the P2P platform,it will amplify its risk and affect the stability of the entire financial system.the incident of Ezubao,Dada Group and ZhongJin have sounded the alarm for the entire Internet finance industry,revealing the significant risks of the P2P network lending platform.Therefore,how to timely and effectively detect the risk of P2P platform has become the focus of discussion in the industry and academic circles.In order to avoid the risks of P2P platform and improve the platform supervision,this paper starts from the specific characteristics of P2P platform and related decisive factors,based on the public information of the Internet,using the big data crawling method to climb the home of online loans,online loans,51 loans,etc.The public data of the third-party website platform,the sample covers large,medium and small platforms,a total of 6424 platform data,around the platform strength,product characteristics,safeguards,risk control capabilities,network services,information disclosure,voter impression in seven aspects,the training set and the test set are divided into a whole,and the dynamic moving window is set.In this paper,we use the four classification algorithm models of logistic model,decision tree,random forest and neural network,and innovatively use dynamic training set and test set method to study the dynamic discrimination of P2P platform risk in China,and horizontally compare the prediction of each model.effect.The results show that the prediction model based on traditional linear regression is generally lower than the model based on machine learning classification algorithm.There is a complex nonlinear relationship between platform default and its related influencing factors.The accuracy of the random forest model is relatively high,and the detection rate of the decision tree model is relatively high,but the cost is also the highest.In addition,the accuracy of the four types of early warning models is relatively stable,with an average of 80%.The detection rate of the model showed a trend of fluctuations and then decreased with time.From 2016 to the first half of 2017,the prediction effect of the model was the best,and the detection rate of the problem platform could reach more than 80%.However,since the second half of 2017,the forecasting effect of the model has declined rapidly,which may be caused by the strictening of the supervision and the concealment of the risk indicators of the problem platform.The key point and difficulty of this paper is to apply data crawler and classification algorithm to the early warning of P2P online loan platform,and build a dynamic early warning system.Based on the empirical analysis results and conclusions,this paper puts forward suggestions for the healthy and sustainable development of China's P2P industry.