Reject Inference In Credit Score

With the rapid development of China’s economy,people’s economic life has undergone tremendous changes compared with 30 years ago,the most obvious of which is the change in consumption patterns.In recent years,credit consumption has gradually become one of the important consumption methods of people.As the main payment medium for credit consumption,credit cards have achieved rapid development since the reform and opening up,which has led to higher and higher requirements for credit risk management.The credit scoring model is an important tool for credit risk management.So for credit consumption,it is important to improve the accuracy of the credit card scoring model.Previous studies on the credit scoring model mostly focused on the credit scoring index system and the credit scoring model.This article focuses on the sample set of credit scores,discusses the characteristics of the sample dataset in the credit score,and analyzes the problem of biased model parameters that may exist in models built on such sample sets.In order to solve this problem,this paper attempts to supplement the missing part of the rejected sample and obtain a relatively complete data set of the sample information before modeling.The method of inferring the default of the rejected sample is Reject inference.This paper first gives an overview of the lack of data,briefly describes the different sample missing mechanisms,and points out that the sample missing mechanism in the credit score problem is an incomplete random missing.Moreover,the mathematical description of the sample with partial loss is carried out.At the same time,it is proved from the mathematical direction that when the sample size is biased,the estimation of the model parameters will be biased,which will affect the accuracy of the model.This also proves that the problem of partial loss of the sample cannot be ignored,and it is necessary to infer the rejected sample and fill in the missing information.Based on the previous studies,this paper proposes a new semi-supervised learning method-CBIL.This paper combines the clustering method with the iterative idea,and uses the prior information of the class to propose the CBIL method to infer the rejection sample,and expounds the clustering thought and iteration conditions of the CBIL method.A Since the default of rejected samples cannot be obtained,this paper first evaluates the inferred method of CBIL through simulation experiments,including accuracy,stability and applicability.The evaluation of the model not only relates to whether the CBIL method can improve the accuracy of model prediction on the basis of logistic regression,but also relates to other methods of rejecting inference to judge whether CBIL has its unique advantages.Finally,through the empirical analysis of the Lengding Club loan data of the United States,the empirical test of CBIL shows that CBIL can still effectively reject the inference and improve the prediction accuracy of the model.