关键词: 线性回归, 线性判别分析, 属性选择, 子空间学习, 稀疏学习

Abstract: The traditional regression model does not ensure to output good performance since it conducts feature selection without considering the correlation between labels. To address this issue, this paper proposes a novel robust low-rank feature selection method. Specifically, this paper considers the correlation between labels into a low-rank regression model and then employs an [l2,p-]norm regularization term to conduct feature selection. Meanwhile, this paper also considers subspace learning method（i.e., Linear Discriminant Analysis（LDA）） into the proposed feature selection model to adjust the result of feature selection. The iteration between feature selection and LDA enables to output optimal features until the algorithm converges. The experimental results on six public datasets show that the proposed feature selection method outperformed four comparison methods.

Key words: linear regression, linear discriminant analysis, feature selection, subspace learning, sparse learning