Supervised Dimension Reduction Through Linear Projection

Abstract

This paper proposes two linear projection approaches for supervised dimension reduction using only the first and second-order statistics for the binary classification problem. They are derived under the general Gaussian model by maximizing the Kullback-Leibler divergence between the two classes in the projected sample. They subsume existing linear projection approaches developed under simplifying assumptions of Gaussian distributions, i.e., when these distributions share an equal mean or covariance matrix. Experiments are conducted to validate the proposed solutions and demonstrate their effectiveness for supervised dimension reduction.