Study on feature select based on coalitional game

Feature selection is an important processing step in machine learning. Most used feature selection methods choose top-ranking features without considering the relationships among features. In this paper, the signification of feature selection is introduced, and the goal and evaluation criteria of feature selection are analyzed. The coalitional game theory related to the feature selection is explained. An algorithm of coalitional game based feature selection (CGFS) is presented. This work focus on selecting a sub-feature set in which the selected features are coalitional and relevant in order to obtain better classification performance. The experimental results show that CGFS obtains better performance than MI.