On A Variable Selection Method based on the Relationship between Discrimination Information and Principal Components

Abstract

Discrimination methods have been proposed by developing a discriminator used with data contains discrimination information. However, when the observed data is multivariate data the discrimination becomes more difficult. In order to solve this problem, a variable selection is important to improve the correct identification rate. In this paper, a new variable selection method is proposed through the principal components by using the relationship between discrimination information and principal components. Several numerical examples show a better performance of the proposed method.