Unsupervised feature selection: A fuzzy multi-criteria decision-making approach

Feature selection (FS) has shown remarkable performance in decreasing the dimensionality of high-dimensional datasets by selecting a good subset of features. Labeling high-dimensional data can be expensive and time-consuming as labeled samples are not always available. Therefore, providing effective unsupervised FS methods is essential in machine learning. This article provides a fuzzy multi-criteria decision-making method for unsupervised FS in which an ensemble of unsupervised FS rankers is utilized to evaluate the features. These methods are aggregated based on a fuzzy TOPSIS method. This is the first time a fuzzy multi-criteria decision-making approach has been used for an FS problem. Multiple comparisons are made to show the optimality and effectiveness of the proposed strategy against multiple competing FS methods. Our approach regarding two classification metrics, F-score and accuracy, appears superior to comparable  strategies. Also, it is performing so swiftly.