以兩階段分類法建構信用卡授信決策模型的實務評估

Empirical Evaluation of Two-Stage Classification Methods on Credit Card Approval System.

巫天虹(Tien-Hung Wu)

指導教授 ： 陳景祥

淡江大學/商管學院/統計學系碩士班/碩士(2013年)

https://doi.org/10.6846/TKU.2013.01030

摘要

近年來，信用卡業務的成長相當的快速，但對於信用卡審核並不嚴謹，使得風險提高，導致2005年底爆發了卡債風暴，使得金融機構承受莫大的損失。
信用卡審核根據申請人過去的信用表現、個人資訊進行判斷，但這些資訊是相當龐大的。因此本研究以兩階段的模型進行預測，首先以F-score法、主成分分析法對眾多的變數進行篩選，將較具有意義的變數保留，並結合邏輯斯迴歸、隨機森林、支援向量機、C4.5、C5.0五種分類器建置模型，進而計算平均預測正確率、敏感度、特異度，進行分類模型評估並與原始未做變數選取的模式進行比較，結果顯示以兩階段方法所建立的模式較原始模式的分類能力能有所提升，並且因變數的減少使模型的運算效率增加。

關鍵字

變數選取 ； 信用卡 ； 邏輯斯迴歸 ； 隨機森林 ； 支援向量機 ； C4.5 ； C5.0

並列摘要

The credit card market has been growing rapidly in recent years but the careless authorization of credit cards made the risk of banks increased. Card debt crisis was occured in 2005 and the banks at Taiwan suffered great loss.
Credit card approval relies on past credit performance and applicant's personal information, but the amount of information is quite large. In this study, we establish prediction models of approval classification by two-stage methods. First, important attributes are selected by F-score and principal component analysis, combined with five different classifiers which are logistic regression, random forest, support vector machines, C4.5 and C5.0, to establish approval models. The average accuracy, sensitivity and specificity of each approach are compared in combination with different classifiers. Our study shows that the two-stage model is better than original classification methods. Reduction of the variables also enhance the computational efficiency.

並列關鍵字

feature selection ； credit card ； logistic regression ； random forest ； support vector machine ； C4.5 ； C5.0

參考文獻

2.李逢嘉(2010)。“特徵選取為基礎之複合分類預測模式-以信用資料為例”.清華大學工業工程與工程管理學系學位論文 2010 年。

1.Berkson, J. (1944). Application of the logistic function to bioassay. Journal of the American Statistical Association, 39(227), 357-365.

3.Cortes, C., &Vapnik, V. (1995). Support-vector networks. Machine learning, 20(3), 273-297.

4.Chen, Y. W., & Lin, C. J. (2006). Combining SVMs with various feature selection strategies. Feature Extraction 315-324. Springer Berlin Heidelberg.

5.Chen, W., Ma, C., & Ma, L. (2009). Mining the customer credit using hybrid support vector machine technique. Expert Systems with Applications, 36(4), 7611-7616.