Variable Screening in Multicategory Classification

• Activity Number: 90
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 8:30 PM to 9:15 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Yue Zeng* and Hao Helen Zhang and Ning Hao
• Companies: The University of Arizona and The University of Arizona and The University of Arizona

Abstract

Classification with high-dimensional features are commonly encountered in many scientific problems in biology, genetics, medicine, and so on. When the number of features is ultra high, a fast and effective dimension reduction is needed or desired to capture important signals, filter out noises, and down-scale the data set without information loss, before a refined and more computationally expensive analysis. In this paper, we study the problem variable screening in multicategory classification problems. A variety of screening procedures are considered, including likelihood-based and LDA procedures, along with screening methods based on pairwise classification. These tools are thoroughly evaluated and compared at various scenarios, and then applied to cancer classification.