Research On Credit Risk Assessment Of Credit Card Customer Based On Two-stage Classification

At present,credit card business is in a rapid-developing period,per capita card and total consumer credit are still at a low level,so there is large space to explore in this area.For credit card applicants,need commercial bank need to check seriously and make accurate credit risk assessments,based on limited customer information.Especially in China,the proportion of customers with good credit history is not high,the majority of the applicant's credit information is very limited.So the critical problem is how to use these limited personal information to assess the applicant's credit risk accurately,and determine his credit level.In this paper,we devised a combination model to determine the credit risk of credit-card applicants based on this credit card business issue.This model consists of a two-stage credit risk identification model and a credit rating index system.The credit risk identification model combine unsupervised clustering algorithm with supervised classification algorithm.Firstly,we use clustering algorithm to divide customer based some properties for the first time,the purpose is to put these customers with common characteristics together;then we`ll use left properties and design its classification individually in every subdivided customer groups.In this paper,,we use three algorithms: the C5.0 decision tree algorithm,the logistic regression algorithm and support vector machine to compare and analysis in the second stage classifier construction.Finally,we use real credit data to verify the results,summarize all classifiers' results and introduce classification error cost matrix in Chapter IV.We compare and assess the classification results in this model with the results using three classification algorithms to the whole data,to prove that "k-means + C5.0 combining algorithm" has the best classification results in building credit risk identification model.Then we'll evaluate customer's credit level according to the probability of default.We'll get customer credit ratings through learning FICO score model based on the repayment ability and willingness to repay in chapter five.The repayment ability accounted for 70%,the willingness to repay accounted for 30%.Finally we divide 100 credit score into 10 grades,and use the credit rating results in credit risk identification model,then we construct a 10 × 10 credit risk judgmental matrix for credit card applicants.Through this judgmental matrix,we can distinguish the customer in detail,different customer credit limits,credit period and interest-free period.So we could find out more customers in the future to bring profit to the bank to make profit maximization while reducing the risk to the greatest extent.