Hyperplane Alignment: Its Implementation, Application, and Advantages

• Activity Number: 303
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Andreas Artemiou*+ and Bing Li and Lexin Li
• Companies: Michigan Technological University and Penn State University and North Carolina State University
• Keywords: sufficient dimension reduction ; Support vector machines ; elliptically distributed predictors ; Inverse regression ; robustness

Abstract

Hyperplane alignment (HA) is a new method for sufficient dimension reduction which can effectively extract linear and nonlinear features in the predictor. In this presentation we focus on the implementation and advantages of this method for linear feature extraction through simulation experiments and real data analysis. Since Hyperplane alignment is based on support vector machine instead of inverse sample moments, it is more robust than traditional dimension reduction methods both against outliers and against non-elliptical distribution of predictors. Furthermore, we demonstrate that it performs well for categorical predictors. An example using the localization sites of proteins in E. coli cells will be discussed