Identification of the Subgroup Signatures by CID

• Activity Number: 363
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: International Biometric Society
• Author(s): Li-yu Daisy Liu+ and Ya-Chun Hsiao* and Tzu-Hao Ma
• Companies: National Taiwan University and National Taiwan University and National Taiwan University
• Keywords: coefficient of intrinsic dependence ; feature selection ; classifier ; microarray

Abstract

Feature selection is the key to the success of accurate classification. Conventional methods of feature selection usually base on statistical hypotheses testing and are thus limited to specific occasions. In this study, we propose the coefficient of intrinsic dependence (CID), which promotes to simultaneously find related features in a broad sense. In addition, CID can be naturally applied to search for subgroup signatures. The above features of CID are particularly important when analyzing complicated and noisy data like microarray expressions. In this study, we first compare CID with conventional statistical methods in simulations. We also apply CID to analysis of breast cancer microarray data. The results demonstrate that CID is a powerful tool to select relevant features, either globally or locally, and greatly improve the accuracy of classification.