摘要
目前的信用卡信用风险研究主要是如何提高模型的预测准确率。针对银行信用卡数据的异质性和信用数据的高度非线性,本文提出了对持卡人信用风险管理的混合数据挖掘方法。该方法包含两个阶段,在聚类阶段,样本数据被聚成同质的类,删除孤立点,不一致样本点重置标签,使样本更具有代表性;在分类阶段,基于样本进行训练生成支持向量机分类器法,对待分样本分类。基于实际数据进行了数值实验,并根据各类样本的特点提出了相应的风险管理策略。
How to increase the accuracy of the forecast model has been a hot issue in the credit risk management. Due to the heterogenicity and non-linearity of the credit data, a hybrid data mining technique combing SOM cluster and SVM classifier is proposed in the paper. There are two phrases in the research：in the clustering phrase, the samples are grouped into homogeneous clusters,and the isolated samples are deleted and inconsistent samples are relabeled. In the classi- fication phrase, the scoring model has been＇built by the support vector machines with samples of new labels. Then experiment is done using the credit data provided by a local bank, and risk man- agement strategies are developed according to the characteristics of data.
出处
《财经理论与实践》
CSSCI
北大核心
2012年第5期36-40,共5页
The Theory and Practice of Finance and Economics
基金
湖南省社科基金(2010YBB127)
湖南省金融工程与金融管理研究中心2010年度开放基金课题(10FEFM02)
湖南科技大学博士科研启动项目(E51022)
教育部人文社会科学研究项目部支持项目(11YJA630124
12YJA630081)
关键词
信用风险
风险管理
数据挖掘
聚类
支持向量机
Credit Risk
Risl
Management
Date Mining
Cluster
Support Vector Machines
