基于决策树信用卡风险管理的研究

Research on Credit Card Risk Management Based on Decision Tree

自从信用卡引入我国以来,其发展速度与日俱增。伴随着我国经济的发展和国民消费习惯的改变,信用卡业务将成为人们生活中不可或缺的一部分。为了让信用卡更好地为经济发展和人们生活服务,信用风险管理就显得尤为重要。因为信用风险是信用卡业务的主要风险。所以发卡机构应在信用风险管理方面加强控制,建立一套合理有效、科学完善的信用评价体系。一般情况下,信用评价会将客户划分为'好'或者'坏'客户。根据客户的个人信息和历史数据,通过数学方法进行建模,进而来预测信用卡客户的违约风险。在这方面,决策树算法正确率高,而且直观简单。本文使用决策树作为建立信用评价模型的方法。本文首先对涉及的文献进行了整理归纳,将信用卡风险管理理论、决策树的理论框架进行结合,分析了决策树、信用卡业务中信用风险的特征。采用决策树C4.5、C5.0、rpart、xgboost、组合预测模型算法进行建模,并且和支持向量机进行了比较。采用决策树方法构建的个人信用评价模型具有精度高、操控性强,并且在实践中使用广泛等诸多优点。

Since the introduction of credit cards in China,its development speed is increasing.With the development of China’s economy and changes in national consumption habits,credit card business will become an indispensable part of people’s lives.Credit risk management is particularly important in order to make credit cards better for economic development and people’s lives.Because credit risk is the main risk of credit card business.Therefore,the issuer should strengthen the control of credit risk management,the establishment of a reasonable and effective,scientific and sound credit evaluation system.In general,credit ratings will divide customers into'good'or'bad'customers.According to the customer’s personal information and historical data,through mathematical methods for modeling,and then to predict the credit card customer default risk.In this respect,the decision tree algorithm is accurate and intuitive.This paper uses decision tree as a method to establish credit evaluation model.This paper first summarizes the literatures involved,combines the credit card risk management theory and the theoretical framework ofdecision tree,and analyzes the characteristics of credit risk in decision tree and credit card business.The decision tree C4.5,C5.0,rpart,xgboost and combined forecasting model are used to model and compare with the support vector machine.The personal credit evaluation model constructed by the decision tree method has many advantages such as high precision,strong handling and wide use in practice.