The Research On Credit Risk Influencing Factors Of Consumer Loan Based On SHAP

Under the background of comprehensively promoting consumption and accelerating the construction of a new double-cycle development pattern,consumer loans play an important role in transforming China’s economic development mode,optimizing the economic structure and improving people’s living standards.At present,China’s consumer credit market continues to develop rapidly,so it is necessary to pay attention to the potential financial risks behind it.Moreover,with the tightening of industry regulatory policies,consumer credit institutions need to regulate their own development and improve the ability to control borrowers’ credit risks.This paper mainly focus on commercial Banks failed to cover the long tail of customer group,Light GBM model is established based on consumer credit data set,and introduce the default Light GBM model results are displayed shapes model explain attribution analysis,rich on the influence factor of consumer loan credit risk identification,make up for a lack of the interpretability of machine learning algorithm.And strengthen the visualization of the influence of factors from the perspective of whole and individual,single factor and multi-factor interaction.The results show that,firstly,Light GBM and SHAP model have advantages when combined with attribution analysis in terms of classification performance and feature importance comparison.Secondly,through the importance of features,this paper finds that the borrower’s personal basic information and the platform’s historical loan information are more important to the prediction of consumer loan credit risk,and the time series and the ratio of personal basic in formation rank higher in the importance of features.Thirdly,this paper analyzes the influence of factors.In terms of single factor influence,the borrower’s personal basic information causes a greater change in the borrower’s expected credit risk than other types of factors,and SHAP values of factors such as the number of prepayment times and loan annuity income ratio in the last five instalments show phased trends.In terms of the interaction of multi-feature factors,this paper finds that when borrowe rs take out loans for several times,the probability of default prediction of people with high credit scores will increase,while the probability of default prediction of people with low credit scores will decrease.At the same time,education level,gender,age and other factors interaction is more apparent,for some credit qualification may normally good groups,such as highly educated,highly educated,highly educated married older women,when they choose consumer credit agencies the high cost of the loa n,there may be adverse selection,the expected level of credit risk is likely to increase.Fourth,from the perspective of individual borrowers,their own risk characteristics are quite different.Finally,based on the research results,this paper puts fo rward policy suggestions from three aspects:consumer credit institutions,regulatory agencies and credit investigation system construction.