Cloud Score for Feature Selection

Feature selection has been studied widely in literatures in supervised learning scenarios. Feature selection methods are categorized into two classes: ¿wrapper¿ and ¿filter¿ approaches. In this paper, we propose a filter method called cloud score based on membership cloud model in fuzzy field. We determine the discrimination power of a certain feature by cloud score to evaluate the feature´s importance. This method is compared with variance and Fisher score methods on UCI iris dataset. The results of the experiments demonstrate the feasibility of our algorithm.