A novel framework for multi-label feature selection: integrating mutual information and Pythagorean fuzzy CRADIS

In recent years, there has been a growing interest in multi-label data classification, with a particular emphasis on multi-label feature selection. While various information-theoretic methods have been devised to determine feature correlations, a majority of them rely on sequential and greedy search strategies for feature subset selection. Concurrently, within the realm of multi-criteria decision-making (MCDM), fuzzy-based methods have gained traction due to their versatile capabilities. In this study, contrary to sequential search, the multi-label feature selection problem is formulated as an MCDM problem with features as alternatives by exploiting mutual information between features and labels. Further, the compromise ranking of alternatives from distance to ideal solution (CRADIS) method is extended to Pythagorean fuzzy sets (PFSs) to solve this MCDM problem and rank the features. To validate the efficacy of the proposed method, a comparative analysis is conducted against six existing methods across twelve benchmark datasets, assessing performance through six multi-label learning evaluation metrics. Furthermore, the proposed approach’s efficiency is proved by the demonstration of statistical significance and stability.