由於台灣地區信用卡的發卡量逐年增加，而且不管是簽帳金額、預借現金或是循環信用金額都逐年快速提高，因此信用卡市場成為金融業的必爭之地。各家金融業者，不論是外商銀行或是本國企業，無不全力搶食這塊大餅。截至94年1月底為止，循環信用金額已高達4,600多億元。
雖然上述金額之利息收入非常可觀，若以20%之循環利率而言，約略年收入將近有920億元；每月的呆帳金額卻高達將近30億元，一年約略有360億元；每月之逾放比率則在3%之普，催收之逾放比約略為每月138億元；若扣除人事成本費用、催收費用及呆帳金額，金融業者的利潤則是大幅縮水。若是利率隨著降低之時，那麼真的是所剩無幾。
因此如何找出良好之客戶，並篩選出信用不良之客戶已經是金融業之重要工作；所以制定一套良好的授信政策及使用正確的分析方法是不可或缺的。本研究運用類神經網路與逐步回歸分析授信相關之變數後發現，如果使用過少之變數，則會造成學習範例及測試範例的錯誤比例相當高；另外當學習筆數越多之時，訓練範例及測試範例正確率相對的提高許多。我們相信在目前的授信政策之下，若能搭配類神經網路做二次篩選，應該可以大大提高不良客戶之過濾，並提高金融業者之獲利。

This project was aimed to apply Neural Networks to enhance factors analysis on credit granting decisions. It is imperative for every financial institution in Taiwan that issues credit cards to attract the customers with decent credits and identify those with poor credits. In order for a credit card issuing institution to become profitable and competitive in Taiwan, establishing viable yet discreet credit-granting policies along with accurate credit analysis is indispensable all along. By applying Neural Networks in factors analysis during credit granting process, it will certainly help each credit card issuing institution make wise yet profitable decisions on granting credits.
The credit card market in Taiwan is booming rapidly over the last several years. The instances on credit card purchases, cash advance with a credit card and card debts in Taiwan continue to stack up at an accelerating rate. For example, the total amount of card debts incurred, as of January 2005, had already hit NT$460 Billion, and counting. It is easy to understand why every financial institution in Taiwan is eager to get into this credit card business. Almost all of banks, local or foreign, in Taiwan have been actively and aggressively taking on this business opportunity in the credit card market.
The main profit gained by any credit card issuing institution comes from the debts accrued through compounding interests. The total amount of the credit card delinquency in Taiwan was NT$138 Billion, as of January 2005. This amount represented a 3.0% of delinquency rate. Consequently, a 20% of APR would have resulted into NT$92 billion in a revenue created by the outstanding debts of credit card as a whole. Under the current circumstances, the bad debts would represent a financial burden which is estimated to be at least 3 Billion NT dollars monthly and 36 billion NT dollars annually.
The revenue generated from credit card interests is enormous by any means. Nonetheless, the profit of a credit card issuing institution tends to be significantly trimmed down by administrative costs, such as personnel and collection expenses, and bad debts assumed by the company. If interest rates ever go down, any single credit card issuing institutions would have gained even less and less profits with such a rate drop. As a result, in order to stay competitive yet still profitable in the market, all of Taiwan’s credit card issuing institutions ought to strengthen the credit granting process and make sound credit decisions that would ultimately lead to financial prosperity.
This project has demonstrated that applying Neural Networks will enhance the credit granting process for each credit card issuing institution in Taiwan. As Neural Networks and Stepwise Analysis were being utilized, their variables showed that the more learning examples and testing examples were implemented, the greater accuracy of both examples would achieve. On the other hand, the lesser examples of both were employed, the less accurate both examples would become. The correlations depicted on the accuracy of learning and testing examples and the impact of Neural Network were evident and positive. It is proven that any credit card issuing institution in Taiwan would be benefited by applying Neural Networks during credit granting process so as to make profitable decisions on granting credits.

1. 財團法人聯合信用卡處理中心網站：http://www.nccc.com.tw/。
2. 中華民國財政部金融局網站：http://www.boma.gov.tw/。
3. 金融聯合徵信中心網站：http://www.jcic.org.tw/。
4. 元大京華證券網站：http://riskmgmt.yuanta.com.tw/models/models/crm_models.htm。
5. 王士峰，王士紘編著，商業自動化，三民書局，1996年，pp.229。
6. 專家系統跨國移轉所面臨之問題研究－以銀行放款評估為例，陳嫦娥82。
7. 葉怡成編著，類神經網路模式應用與實作，儒林出版社，2004年。
8. 羅華強編著，類神經網路－MATLAB的應用，清蔚科技，2001年。
9. 蒙以正，MATLAB5專業設計技巧，碁峰資訊，1998年。
10. 葉怡成編著，應用類神經網路，儒林出版社，1997年。
11. 秉昱科技，模糊邏輯與類神經模糊在商業和財政的應用，儒林出版社，1998年。
12. 財務金融研究中心，投資分析+MATLAB應用，全華科技圖書股份有限公司，2000年。
13. Hagan Demuth Beale，類神經網路設計，Thomson，2004年。
14. 郭敏華，債信評等，智勝文化事業公司，2000年。
15. 彭慧雯，建構信用卡資料挖礦架構及其實證研究，生產系統工程與管理研究所碩士學位論文，2001年。
16. 潘文超，以遺傳演化類神經網路建構e化預測系統－樂透與財務預警個案研究，經濟學系碩士論文，2003年。
17. 江惠櫻，商業銀行對企業授信決策考量因素與授信品質之關係，企業管理學系碩士論文，2002年。
18. 林芝儀，應用資料探勘於信用卡授信決策模式之實證研究，工業工程與管理學系碩士論文，2003年。
19. 許峻源，類神經網路與MARS於資料探勘分類模式之應用，應用統計學研究所碩士論文，2001年。
20. 林勉今，消費性貸款授信風險評估之研究－以X銀行為例，事業經營研究所碩士在職專班碩士論文，2004年。
21. 李美笑，信用卡持卡人信用風險之研究，保險學系碩士論文，2002年。
22. 丁正中，消費金融信用風險研究－信用評分概述，聯合徵信中心風險研究小組。
23. 呂奇傑，演化式類神經網路分類技術於資料探勘上之應用，應用統計學研究所碩士論文，2001年。
24. A.B. Emel et al. / Socio－Economic Planning Sciences 37，A credit scoring approach for the commercial banking sector，2003，pp. 103–123。