Application of variable selection techniques to a modified SAIS for generating practical scorecards
by Kevin Leung, France Cheong, Christopher Cheong, Sean O'Farrell, Robert Tissington
International Journal of Applied Decision Sciences (IJADS), Vol. 2, No. 3, 2009

Abstract: Selecting better predictive variables is fundamental for scorecards to perform well. This study makes use of a large credit scoring dataset and investigates the application of several variable selection techniques for scorecard development. The scorecards are developed using a statistical technique (logistic regression) and two AI methods (SAIS and AIRS). SAIS, which we previously developed can predict class outcomes accurately and has good classification accuracy which is the percentage of correctly classified data. However, since an unbalanced dataset was obtained, the Gini coefficient which is the main performance measure used in industry and which is insensitive to changes in class distribution needs be used instead. SAIS is modified to generate a Gini coefficient and an investigation of its suitability for practical scorecard development is made. We found that further modifications are needed in order for it to perform as well as logistic regression. Moreover, among the different variable selection techniques used, stepwise regression was found to perform best.

Online publication date: Thu, 20-Aug-2009

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Applied Decision Sciences (IJADS):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com