Unsupervised feature selection method based on sensitivity and correlation concepts for multiclass problems

Feature selection is the problem of eliminating the features which are irrelevant and/or redundant. It can also be assumed as the problem of selecting a small subset of features which are necessary and sufficient to describe the target concept. In this paper, a new feature selection method based on the concepts of sensitivity and Pearson’s correlation is introduced which is called Sensitivity and Correlation based Feature Selection-SCFS. The sensitivity of one feature is computed via applying the subtractive clustering and is utilized as feature-target relevancy. Pearson’s correlation coefficient is used to determine the redundancy among a subset of selected features. The introduced measure increases the score of a selected feature subset which has maximum relevancy to the target concept and minimum redundancy among features. The proposed criterion is employed as the fitness function in a genetic algorithm in order to evaluate feature subsets. Some well-known benchmark datasets are utilized for investigating the performance of the proposed method. Also, the results of our method are compared with the other similar feature selection methods. The obtained results show however SCFS is an unsupervised filter; it is well comparable to the other well-known supervised methods in terms of classification accuracy and the number of selected features.