Unsupervised Feature Selection Based on the Distribution of Features Attributed to Imbalanced Data Sets

Since dealing with high dimensional data is computationally complex and sometimes evenintractable, recently several feature reduction methods have been developed to reduce thedimensionality of the data in order to simplify the calculation analysis in various applications suchas text categorization, signal processing, image retrieval and gene expressions among manyothers. Among feature reduction techniques, feature selection is one of the most popular methodsdue to the preservation of the original meaning of features. However, most of the current featureselection methods do not have a good performance when fed on imbalanced data sets which arepervasive in real world applications.In this paper, we propose a new unsupervised feature selection method attributed to imbalanceddata sets, which will remove redundant features from the original feature space based on thedistribution of features. To show the effectiveness of the proposed method, popular featureselection methods have been implemented and compared. Experimental results on the severalimbalanced data sets, derived from UCI repository database, illustrate the effectiveness of theproposed method in comparison with other rival methods in terms of both AUC and F1performance measures of 1-Nearest Neighbor and Na ve Bayes classifiers and the percent of theselected features.