近年來由於塑膠貨幣的崛起，使得信用卡市場蓬勃發展，並在近十年間成長了近16倍。也使得各家銀行對信用卡市場莫不投入相當的資源以搶食這塊大餅。然而，也因為競爭激烈，使用各家銀行莫不降低申請的門檻，以提高客戶申請的意願。然而，也因為銀行忽略了發卡的風險管理，使得銀行因卡債所產生的呆帳年年增加。產生了未蒙其利，先受其害的現象。所以，如何在發卡之前，針對客戶的歷史資料，建構出一套有效的預警系統，降低銀行的損失，為本研究的目的。

本研究根據銀行的資料主檔、信用卡資料及客戶的聯徵中心資料等，應用類神經網路分群的特性，找出信用不佳的潛在因素，並利用關聯度的方法，找出與信用不佳關聯較高的變數。藉以預測潛在信用不佳的客戶，而實證結果亦顯示，預測的準確率達九成以上。隨著預警系統的準確率提高，相信亦能讓銀行在客戶的風險管理上更能確切的掌握。

本研究著重於針對銀行現存客戶進行預測，未來若能採集新客戶的資料樣本作研究，找出新客戶是否有某些特徵可依循，相信未來亦可針對新客戶作風險預測，以使銀行降低發卡風險，提高獲利。

Due to the use of plastic currency rising abruptly in recent years，the credit card market grow up vigorously and also it increased near 16 times in the past 10 years. All of banks devoted to make a large investment in the credit card market in order to get the maximum profits in the competition market. However， in this intense competition market，all the banks are trying to reduce the requirements for customer application in order to enhance the customers’ intention to apply credit card. Because of that， the banks ignored the risk management of credit card approval that makes the bad debt increased every year. That is why the bank did not get the profits but a great loss. The object of this research is how to build the warning system which base on the information of customer history to reduce the banks’ losses before the credit card approval.
The research is according to the database of banks，including main data、credit card data and the customer data from JCIC(Joint Credit Information Center), and then use the characteristics of neural network to find out the potential factors of the bad credit and finally use relation method to find out the higher (important) relative variable (parameters) of bad credit. The study results show that the accuracy of forecast rate of the warning system is 90% above. Along with early warning system''s rate of accuracy enhancement，the bank can also regulate the risk management of customers more easier.
The research is focus on forecast of the existence customers of the bank. If the collection of new customer data are available，we believe the warning system will work well and the bank will lessen risk of card approval and will increase the profits.

目錄
書名頁……………………………………………………………………i
審定書………………………………………………………………… ii
授權書…………………………………………………………………iii
中文摘要……………………………………………………………… iv
英文摘要……………………………………………………………… v
誌謝……………………………………………………………………iv
表目錄 ix
圖目錄 x
第一章、緒論 1
1.1研究背景與動機 1
1.2研究目的 4
1.3論文架構 5
第二章、文獻探討 7
2.1信用卡的風險 7
2.2信用貸款評估 9
2.3資料探勘 11
2.4文獻回顧 12
第三章、研究方法 18
3.1資料來源 18
3.2工具軟體 18
3.3分群技術 20
3.4灰關聯分析 24
第四章、實證結果與分析 26
4.1銀行判斷信用不佳的依據 26
4.2資料庫整理與變數分類 28
4.3變數定義 32
4.4以類神經網路模型預測持卡人信用不良的因素 37
4.5分群結果分析 38
4.6預警系統建構 43
第五章、結論與建議 46
5.1研究分析與結論 46
5.2未來改進方向 49
參考文獻 50
附錄(一) 52
附錄(二) 54
附錄(三) 55
附錄(四) 56
附錄(五) 57
附錄(六) 58
附錄(七) 59

[1].陳俊源(2003)，「T-SQL指令語法與預存程式徹底研究」，初版。
[2].陳銘憲(2006)，「鑑往知來：談智慧化的基礎技術－資料探勘」專題演講。http://www.nsc.gov.tw/_newfiles/head.asp?add_year=2006&tid=53
[3].陳宗文(2001)，「信用卡貸款」，金融與徵信叢書No.45，初版。
[4].陳炳良(1999)，「信用卡的風險管理」，金融財務季刊，第三期，頁77~91。
[5].徐素芬，淺談信用卡詐期偽冒風險及其管理，中信通訊，頁19~22。
[6].林炯垚(1997)。金融創新與風險管理，初版。
[7].鄧聚龍，郭洪(1996)，灰色原理與應用。
[8].張智星，Data Clustering and Pattern Recognition. http://140.114.76.148/jang/books/dcpr/3.3-kMeans.asp?title=3-3%20K-means%20Clustering
[9].詹書銘(2005)，「演化式類經網路、灰關聯分析應用於信用卡風險管理之實證研究」，朝陽科技大學財務金融系碩士論文。
[10].王信勝(2001)，「整合分析層級程序與類神經網路之信用評分模型」，輔仁大學資訊管理學系碩士論文。
[11].張勝春(2001)，「模糊類神經在銀行授信決策之應用」，朝陽科技大學財務金融系碩士論文。
[12].彭美珊(1994)，「授信決策應用模糊理論之研究」，台灣經濟第206期。
[13].陳慶瀚（2002），「類神經網路」，義守大學電機系課程講義。
[14].金融聯合徵信中心個人信用評分簡介，聯合徵信中心網站，http://www.jcic.org.tw/index.htm
[15].T.Khabaza and C.Shearer(1995)，“DATA MINING WITH CLEMENTINE”，integral Solutions Limited.
[16].A.Reitano,F.D.Maio, and S.Seminara(2005)，“Data Mining and Management Decisions”,International Conference on Computer Systems and Technoloyies.
[17].G.Adomavicius and A.Tuzhilin(2001)，“Using Data Mining Methods to Build Customer Profiles”，IEEE.
[18].Su,Mu_Chun and Hsiao_Te Chang（2000），“Fast Self-Organizing Feature Map Algorithm.”IEEE Transaction on Neural Network.
[19].Ritter,H.J. and T.Kohonen(1989)“Self-Organizing semantic maps,”Biolog.Cybern(61).
[20].McCallum, A.K.Nigam, and L.Ungar.(2000)“Eifficent clustering of high-dimensional data sets with application to reference matching.”In KDD.
[21].Miyoshi,T.,H.Kawai and H.Masuyama(2002)“Efficient SOM Learning by Data Order Adjustment”Neural Networks, IJCNN ’02.Proceedings of the 2002 International Joint Conference(1).
[22].Rosenbloom, J.S.（1972）,“ A case study in risk management”，NY：Meredith Corp.
[23].Williams, C.A. and Henis, R.M.(1981), “Risk management and Insurance”，4th Ed.,NY:McGraw-Hill.
[24].Orgler,Y.E.(1970),“A Credit Scoring Model for CommercialLoans,Journal of Money,credit,and Banking.”
[25].Capon,N(1982),“Credit Socring System：A Critical Analysis”,Journal of Marketing.
[26].Srinivasan, v. &Kim, YH.(1988),“Designing Expert Financial System A Case Study of Corporate Credit Management”,Financial Management.
[27].R.A.Neibet，”How to Choose a Data Mining Suite”，http://www.dmreview.com/article_sub.cfm?articleID=1000465,2004