Centroid Shifted Linear Discriminant Analysis for Semi-supervised Learning
Yong-Joon Shin, Cheong-Hee Park · 2011
Semi-supervised dimension reduction can be used when the size of labeled data samples is small. In this paper, we propose a semi-supervised dimension reduction method, called centroid shifted linear discriminant analysis(CSLDA), which performs linear discriminant analysis after shifting class centroids using unlabeled data samples. Experiments using high dimensional text data demonstrate that the proposed method can improve classification performance greatly.