An Integrated Feature Selection and Rough Sets Approach for Classifying Credit Ratings in Asian Banking Industry

• Main Authors: Yan-Yu Chen, 陳彥宇
• Other Authors: Ching-Hsue Cheng
• Format: Others
• Language: en_US
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/26204846797385372464

碩士 === 雲林科技大學 === 資訊管理系碩士班 === 97 === Although Asia is the most energetic and expanding economy, it connects with high investment risks and uncertainty. Credit ratings provide an objective opinion in assessing the credit worthiness, investment risks, and default probability of issue or issuer, and they are benefit to investors, issuers, regulators, and other stakeholders. This thesis proposes a new procedure to classify credit ratings, analyze credit ratings determinants, and provide meaningful credit ratings rules in Asian banking industry. Firstly, an integrated feature selection approach is employed, including: (1) Chi-Squared, (2) ReliefF, (3) Gain Ratio, and (4) InfoGain. Secondly, we use an objective method—CPDA (cumulative probability distribution approach)—to partition condition attributes by using rough set local-discretizition cuts. Thirdly, rough sets LEM2 algorithm is employed to generate decision rules. Finally, in order to improve rule quality we filter rules that have lower support. The experimental dataset retrieved from Bankscope database includes 1327 banks in Asia. The experimental result shows that the integrated feature selection approach is an effective way to remove irrelevant attributes, and the proposed procedure provides a precise and comprehensible classification rules.