Research On Consumer Loan Default Risk Prediction Based On XGBoost Model

In the context of rural revitalization,the use of consumer finance to sink to various rural areas,and the use of consumer finance to meet the consumption needs of farmers in rural areas,has become an important choice to promote rural revitalization and development.For rural areas,rural commercial banks are the core institutions in rural finance at present.Agricultural and commercial banks are also increasing the development of consumer finance to promote the consumption level in rural areas.At present,Rural Commercial Bank and other banks mainly judge the default risk of consumer finance based on various scoring rules,which is difficult to further improve the prediction effect.In data mining,XGBoost and other models have been well applied in various scenarios due to their good prediction ability.This paper studies the default risk prediction of individual consumer loans of agricultural and commercial banks.First of all,taking A Rural Commercial Bank as the specific research object,through the analysis of various data of the bank’s consumer finance and default,this paper summarizes the problems faced by Agricultural Commercial Bank in terms of consumer finance default risk and other aspects.Secondly,based on the "5C" theory and combined with the characteristics of consumer loans of agricultural commercial banks,the prediction system of consumer financial default risk is constructed.Thirdly,based on XGBoost,build the prediction model and process,and use particle swarm optimization to optimize the parameters.Fourth,based on the sample of A Rural Commercial Bank’s personal consumer finance,this paper elaborates on the treatment of the sample,the realization of the specific default model,the prediction effect on the sample and the test situation.Finally,some suggestions are put forward to improve the prediction ability of agricultural commercial banks to default risk.The research conclusions of this paper are as follows:(1)Taking A Rural Commercial Bank as the specific research object,through the analysis of various data of the bank’s consumer finance and default,the comprehensive analysis of this chapter can see that the indicators used by A Rural Commercial Bank to judge consumer lenders are not comprehensive enough,and the accuracy of the evaluation method is not high;(2)From the comparative analysis of different models,XGBoost has better performance in the test set of personal consumer loans.Compared with the comprehensive scoring method of Agricultural and Commercial Bank of China,the accuracy rate and other indicators have been improved,with an overall increase of 0.1006 in AUC.XGBoost has better performance through cross-validation with 10% discount.Compared with the comprehensive scoring method of Agricultural and Commercial Bank of China,the accuracy rate and other indicators have been improved,with an overall increase of 0.1051 in AUC;(3)Based on the index set constructed in this paper,the prediction is carried out.Compared with the original prediction index set of ABC,the AUC is increased by 0.0287,and the cross-validation is increased by 0.0367.The debt-income ratio has a high impact weight on the prediction of the default risk of personal consumer loans.At the same time,the judicial litigation information also helps the agricultural commercial banks to judge the default risk of lenders.The credit information of family members also has a certain impact on the predicted default,indicating that the credit risk between family members may be transferred to a certain extent.