An Approach for Building Feature Graphs in feature selection algorithms

Graph-based methods have recently gained significant attention in feature selection due to their ability to effectively capture complex relationships between features through graph topology. Most existing methods rely on graph theory and search techniques to identify high-quality feature subsets. However, the construction of an appropriate feature graph—an essential component for achieving optimal feature subsets—has been relatively overlooked. To address this limitation, this paper introduces a novel mechanism for representing relationships between features, along with a new graph-based unsupervised feature selection method. Additionally, a novel approach for identifying key nodes within the feature graph is proposed. The algorithm begins by constructing the feature graph without requiring specific parameters. It then detects communities within the graph, where the members of each community elect representatives for the final feature subset. Experimental results demonstrate the effectiveness of the proposed algorithm, highlighting the importance of accurate relationship representation in feature selection.

کلیدواژه ها:

Data mining ، feature selection ، graph theory ، classification