Research On Online Loan Credit Risk Model Based On Machine Learning Algorithm

Internet finance has made considerable progress today.The state’s emphasis on this field has gradually increased.In order to better understand customer behavior and profiles,the financial industry has an increasing demand for data mining and analysis.For financial institutions that issue credit,the ability to distinguish between good and bad customers is very important.The improvement of forecast accuracy will help the sustainable and healthy development of these institutions.In addition,early warning of customers with significant default risks can help lending institutions prevent non-performing loans and encourage customers to better manage their personal finances.Online loan is a new type of lending model that relies on Internet media,and it complements the gray areas that are not involved in the traditional financial industry.Among them,whether the user’s credit risk can be accurately assessed is a prerequisite to ensure the normal operation of the relevant financial project or platform.In this paper,the preliminary selection of credit evaluation indicators is to design a comprehensive evaluation index system for borrowers’ online loan credit risk from five first-level indicators such as identity characteristics,credit history,debt solvency,loan product element information,and daily behavior.Adopting "multiple" and "multidimensional" data cleaning methods,applying Light GBM and XGboost two machine learning algorithms to predict the risks of transaction data on the online lending platform,which is innovative in the prediction of loan default risks.In addition,through comparison and hypothesis testing,we have observed that the model fitted by the Light GBM algorithm combined with the multiple observation cleaning method is the best.Finally,in view of the influencing factors of the default rate,it is proposed that the online loan platform should further improve its own measurement indicators from two important aspects: increasing the sample data volume and economic evaluation level.At the same time,it is proposed that the government should actively establish a sound and transparent petition system and its development direction in this field.