Completed sample correlations and feature dependency-based unsupervised feature selection

Abstract Sample correlations and feature relations are two pieces of information that needed to be considered in the unsupervised selection, as labels missing guide model construction. Thus, we design a novel selection scheme, this paper, via considering completed sample dependencies unified framework. Specifically, self-representation graph construction conducted preserve select important neighbors for each comprehensive way. Besides, mutual sparse learning designed consider between features remove informative features, respectively. Moreover, various constraints constructed automatically obtain number conduct partition clustering task. Finally, test proposed method verify effectiveness robustness on eight data sets, comparing with nine state-of-the-art approaches regard three evaluation metrics