| As the world’s second largest economy,China’s social economy has made great development since the reform and opening up,and the current economy is also in a new stage from high-speed development to high-quality development.More and more people in China choose to use credit cards to spend in advance.As an important part of China’s financial industry,the main business income of banking comes from credit loan business.Therefore,how to effectively measure the possible default risk is very important.Through the literature research at home and abroad,it is found that the research on credit default identification in foreign countries started earlier and the method is relatively perfect,while the domestic research started relatively late,and there are some deficiencies in the selection of model establishment indicators.Therefore,it is more and more important to promote and promote the high-quality development of China’s banking industry,improve China’s banking credit management system,and build a credit model suitable for China’s banking industry.This paper processes and analyzes the credit data set published on cosai website from 2015 to 2017,which has 21 variables.First,preprocess the data,and record the overdue and non overdue status of the dependent variable user as 1,0.The classified variables are transformed into dummy variables and processed in the same order.The situation that significantly affects the overdue is recorded as 1.And check whether there are missing values and exceptions in the numerical variables.After that,modeling and analysis are carried out.After comparing and analyzing various credit evaluation models in previous studies,this paper finally selects XGboost model,random forest logistic model,adaptive lasso logistic model,and forward random forest adaptive lasso logistic model.The confusion matrix,accuracy,precision,recall,KS curve,ROC curve and other indicators are selected to compare and analyze the model.Finally,it is found that the random forest adaptive lasso logistic model with the ability of variable selection has better prediction ability of personal credit default risk,and the effect is significantly improved compared with the ordinary logistic model.At the same time,according to the model,we also get the variables that significantly affect personal default,such as loan amount,loan interest rate,loan term,whether it has passed academic certification,video certification,credit certification,historical repayment status and gender.Therefore,the model proposed in this paper can effectively identify and predict the personal credit risk default of financial institutions.Finally,the paper puts forward some suggestions from the perspective of improving relevant policies,regulations and supervision system and strengthening credit risk management of financial institutions,so as to improve China’s personal credit risk system. |
