以資料探勘之決策樹方法建立小額信貸之信用評分模型研究

Appling Data Mining Technique in Building Credit Scording Model for Consumer Loan Appliaction

張慶光(Ching-Kuang Chang)

指導教授 ： 陳文華

國立臺灣大學/管理學院/商學組/碩士(2006年)

https://doi.org/10.6342/NTU.2006.10448

摘要

消費金融的壞帳在近二年呈現高速的成長，尤其雙卡所衍生的卡債及卡奴問題不僅大大影響了金融穩定，亦成為一個嚴重的社會問題。而由於小額信貸提供較雙卡優惠的利率，使得許多卡債已移轉至小額信貸。因此，幾乎可以確定「小額信貸」業務將成為另一個受害者。為減少潛在壞帳的發生，銀行實有必要建立一套科學化的審核機制，來區別好壞客戶。
本研究主要在針對銀行的小額信貸業務，以實際的客戶資料配合先進的資料探勘技術，試圖建立一個「信用評分模型」，以供銀行後續的業務發展參考。本研究建立之信用評分模型經測試資料驗證後其正確預測率高達86.8%；Area Under Curve（AUC）值亦高達84.6%，已具有一定程度的信用風險鑑別能力。茲將研究結論摘錄如后：
1. 信用評分模型的導入，可有效降低違約率，提昇整體利潤
2. 嚴謹的負面表列設定，可補足模型本身的限制
3. 信用評分與人員審核不可偏廢，核貸流程應重新設計
4. 資料探勘方法可建立具客觀性的評分規則
5. 銀行不應盲目追求業績，而應建立最適風險政策
6. 若能建立一信用風險評估機制，則「小額信貸」業務仍為一重要的獲利業務

關鍵字

資料探勘 ； 信用評分 ； 決策樹

並列摘要

The growth in consumer credit outstanding over the last 2 years is truly spectacular. The bad debt issue generated by credit card and cash card business has not only affect the stability of financial system, but also the foundation of our society. It is pretty sure that the Consumer Loan business will become the next victim of non-performing loan, because the balance will be transferred from card business to consumer loan due to the lower interest rate. To reduce the potential default rate, lenders have to establish a mechanism to discriminate between good and bad customers.
The goal of this study is building a Credit Scoring Model, by applying the advanced Data Mining Technology – Decision Tree algorithms, for Consumer Loan Application. The best model was obtained by evaluating both the accuracy rate and AUC value using the testing data which is 30% of the original sample. The result of this research indicate that
1. Implementing credit scoring model can not only reduce the credit risk, but also increase the overall profit of financial institutions.
2. Rigorous filtering rules are essential to complement the insufficient part of model itself.
3. Scoring Model and loan officer’s verification should not be neglected.
4. Impersonal rules to discriminate between good and bad customers can be produced by Data Mining techniques.
5. The ultimate goal of consumer loan should focus on the profit rather than the loan amount.
6. Consumer loan can be a profitable business to banks, if the credit scoring model can be implemented in a right manner.

並列關鍵字

Data Mining ； Credit Scoring ； Decision Tree

參考文獻

20. 探宇科技股份有限公司，「UniCreditTM智慧型信用評分系統」操作手冊，民國94年

6. 林芝儀，「應用資料探勘於信用卡授信評量模式之實證研究」，元智大學工業工程與管理研究所論文，民國92年

1. Altman, E. I., “Managing Credit Risk: The Challenging for the New Millennium”, Stern School of Business, New York University, 2002

4. Beares, P. B., “Consumer Lending”, 3rd Edition, American Bankers Association, 1997

6. Feelder, A. J., “An Overview of Model Based Reject Inference for Credit Scoring”, Utrecht University, 2003