信用评分模型应用比较研究——基于个体工商户数据的检验  

An Comparison Study on the Consumer Credit Scoring Models: Based on the Data of Self-employed

作　　者：晏艳阳[1] 蒋恒波[2] 

机构地区：[1]湖南大学统计学院 [2]湖南大学金融学院,湖南长沙410079

出　　处：《统计与信息论坛》2010年第5期30-35,共6页Journal of Statistics and Information

基　　金：国家社会科学基金项目<我国个人征信中的基础技术研究>(06BTJ020);湖南省软科学重大项目<湖南省财政支出绩效评价体系研究>(2009ZK2007)

摘　　要：信用评分是各类机构进行信用管理的有效工具,有着广泛的应用前景。随着计量技术的发展,信用评分方法也不断革新,为实际应用提供了多种选择。选取Logistic回归、分类树两种统计方法及代表信用评分发展趋势的人工智能神经网络中的多层感知器、径向基网络、自组织特征映射网络、支持向量机等共六种模型,运用较大样本量的个体工商户数据在一致的框架下进行检验。结果表明:Logistic回归模型与支持向量机两种方法在错分率、稳定性及适用性方面较为优越,其中支持向量机作为人工智能评分方法的最新应用之一,其综合性能更为突出。Credit scoring is an effective credit management tools in various institutions,and have broad application prospects.With the development of measurement technology,credit scoring methods have been innovative,provide broad choices practical applications.This paper selects Logistic regression,classification trees and a total of six models of artificial neural network,namely MLP,radial basis function network,self-organizing feature map,support vector machine,which representing the trends in the credit scoring modeling;uses a larger sample size of self-employed business data in a consistent framework to conduct statistical test.The results show that,Logistic regression model and support vector machine are more advantageous in misclassification rate,stability and applicability.In the two of them,as the latest applications of artificial intelligence scoring,support vector machines is the method who has more prominent overall performance.

关 键 词：信用评分模型 统计方法 人工智能神经网络 个体工商户 

分 类 号：O212[理学—概率论与数理统计]