商业智能在信用卡违约户特性分析之应用

Application of Business Intelligence for Pattern Analysis of Credit Card Default Prediction

李建成(Carlos Lee)

《Journal of Data Analysis》 2卷6期 (2007/12) Pp. 191-203

https://doi.org/10.6338/JDA.200712_2(6).0011

摘要

金融改革以来，其中在消费金融商品为最活络的业务，当中又以「信用卡」的推广最为成功。而发卡银行面对申请案件，若只采用人工审件方式判断是否应发放信用卡及其信用额度，将造成人力资源与时间浪费，亦容易因为个人判断标准不一而影响信用审核质量。另一方面，对已持卡者之信用状况的管理亦相当重要，由于信用风险涉及许多层面，因此近年来所被重视的数据挖掘技术就成为其重要的一项工具。若能及早发现可能产生呆帐等违约情形之持卡者，对其行为进行监控，将可有效预防违约的发生。因此，本研究主要目的在于利用商业智能与数据挖掘的技术整合，希冀建立一套相对稳定且有效的预测模型，提供相关部门与发卡机构一个准则，以降低违约比例，进而降低信用风险，藉此提升该银行在市场上的评价。根据本研究所选定罗吉斯回归当作分析工具后，在经过不同的抽样方式，其所建置出来的模型，对于整体预测度、非违约户的预测度、与违约户的预测度皆有七成以上的预测能力，与国内外相关研究团队所建出70%～80%预测度的模型相距不大，显示出本研究的模型具有一定的预测水平，所以在分析工具的选择及抽样的过程，亦可提供给尔后从事相关研究者作为参考。

關鍵字

信用卡 ； 违约户 ； 数据挖掘 ； 罗吉斯回归 ； 类神经网络

並列摘要

Since the financial reform, the promotion of ”credit card” has been the most successful commodity in consumption finance. When the issue card bank deals with applications, the method of artificial examination in judging credit amount and whether if the credit card is issued or not will waste time and human resources. The quality of credit evaluation can be easily affected by the inconsistences differences of individual judgment as well.On the other hand, it is important to manage the credit condition of card owner as well. Owing to the credit risk involves many facts, data mining has become an important tool in recent years. If we could find the probabilistic defaulters earlier, and monitoring their behavior, it might prevent the occurrence of default effectively.Therefore, the main purpose of this research lies in the combined technology of commercial wisdom and data mining. We try to establish a set of stable and predictive model effectively. Provide a standard for the related departments and the issue card organizations. This can reduce the proportion of default and credit risk. By means of the advantages, we can promote the evaluation of the bank in marketing.Our research chooses Logistic Regression as the designated tool, and builds a model after different sampling methods. Our model can predict more than 70% of the general prediction, non-default prediction and default ones. That a domestic and foreign research team builds a 70% to 80% accuracy prediction model shows our model has had certain credits. Therefore, our research can provide related researchers with future references in the selection of analytical tools and the process of sampling.

並列關鍵字

Credit card ； Default Prediction ； Data Mining ； Cross Industry Standard Process for Data Mining ； Logistic Regression ； Neural Networks ； CRISP-DM

參考文獻

韦端、郑宇庭、邓家驹、匡宏波、谢邦昌(2003)。DATA MINING概述—以Clementine7.0为例。中??据挖掘??。

施孟龙、尤清芳、李佳珍(1999)。Logit Model应用于信用卡信用风险审核之研究。金融财务月刊。10月

曾玫苑(2004)。信用卡信用风险评估。私立东吴大学国际贸易学系研究所。

刘泰谷(2004)。信用卡信用评分模型之建构与分析。私立世新大学财务金融研究所。

廖仁杰(2005)。信用卡业务信用评分制度与模型之有效性研究。国立中央大学财务金融学系硕士在职专班。