Class-specific variable selection for multicategory support vector machines

This paper proposes a class-specific variable selection method for multicategory support vector machines (MSVMs). Different from existing variable selection methods for MSVMs, the proposed method not only captures the important variables for classification, but also identifies the discriminable and nondiscriminable classes so as to enhance the interpretation for multicategory classification problems. Specifically, it minimizes the hinge loss of MSVMs coupled with a pairwise fusion penalty. For each variable, this penalty identifies nondiscriminable classes by imposing their associated coefficients in the decision functions to some identical value. Several simulated and real examples demonstrate that the proposed method provides better interpretation through class-specific variable selection while preserving comparable prediction performance with other MSVM methods.

Keywords

fusion penalty, lasso, support vector machine, variable selection

2010 Mathematics Subject Classification

62F07

Full Text (PDF format)

Published 28 February 2011