A Fuzzy Integral Approach for Ensembling Unsupervised Feature Selection Algorithms

Abstract

Feature selection is an effective technique for decreasing data dimensionality by selecting a significant feature set. Gathering label information can be time-consuming and expensive, as labeled instances are not always available. Therefore, unsupervised learning importance has emerged. In this article, a new unsupervised feature selection is presented based on an ensemble strategy. The ensemble of multiple feature selection methods is performed using fuzzy integral operators. The comparisons are made against various feature selection methods in the literature to show the better performance of the proposed method. These comparisons are conducted based on classification accuracy and run-time.