A graph theoretic approach for unsupervised feature selection

Parham Moradi, Mehrdad Rostami

Abstract: Highlights•A novel graph-theoretic approach for unsupervised feature selection is proposed.•Our method integrates graph clustering with a novel iterative search strategy.•A node centrality measure is used to identify representative and informative features.•The size of final feature set is determined automatically.•Our method has been compared to the well-known and state-of-the-art methods.