Application Exploration Of Consumer Finance Credit Scoring System

Accompanied by social consumption level of ascension,the prosperity of the consumer market has promoted the rapid development of the financial credit industry.The credit card business dominated by banks is no longer the only choice for individual users to participate in credit activities.In recent years,consumer finance credit launched by nonfinancial banking institutions has been favored by individuals.Consumer finance credit is a kind of low limit loan product which provides users with life consumption.The consumer finance companies holding consumer finance licenses issued by the CIRC can provide such credit products for users.This kind of products can be applied for authorization by users online,which has the advantages of low single transaction amount,simple audit process,fast payment and so on.Although the huge market brought by the convenience and rapidity of its transaction process can bring huge profits to the credit industry,the reduction of the approval process also brings the problem of low accuracy of user credit risk audit,which is prone to financial credit fraud.So the consumer financial institutions in recent years more and more emphasis on the user's credit risk assessment.Based on the analysis of the performance of the traditional machine learning algorithm in the credit evaluation model,this paper proposes a fusion Gradient boosting forest algorithm.This algorithm integrates the ideas of Bagging ensemble learning and gradient boosting into the construction of credit risk assessment model,and uses it and traditional machine learning algorithm to build the credit risk assessment model respectively.By comparing the prediction results of the model,it is proved that the model has the same prediction ability as the Random forest and XGBoost model,and it is verified that the Gradient Boosting Forest model has good stability in reducing the bias and variance.This paper first summarizes the credit scoring system in domestic and foreign development present situation and the credit evaluation system set up involves the theory of knowledge.Then,the credit score card model is built based on the transaction samples of a financial institution.Firstly,12 variables were selected as model characteristics through the process of exploratory analysis,feature cleaning and feature selection.Then,four algorithms of Logistic regression,Random forest,XGBoost and Gradient boosting forest are selected to build the credit risk assessment model.The prediction results of the model show that the Gradient boosting forest and two external ensemble learning models have achieved good prediction results on the sample set.At the same time,the prediction effects of Gradient boosting forest,random forest and XGBoost in the test set are discussed from the perspective of variance and bias respectively,and the stability of Gradient boosting forest algorithm in reducing bias and variance is further demonstrated by comparing the results.Finally using Logistic regression and Gradient boosting forest model in predictive results of the test set for the generation of credit score card,by discussing the differences of two credit score card results,the results show that the Gradient boosting forest algorithm accuracy and degree of differentiation is superior to the traditional Logistic regression models,further discusses the effectiveness of the proposed algorithm.This exploration in the field of consumer finance credit evaluation provides a new idea for the development of financial industry in the direction of financial risk control.