Intuitionistic Fuzzy Rough Mutual Information Aided Feature Subset Selection and Its Applications

Abstract

Feature selection plays an essential role in solving the challenges of the "curse of dimensionality" in data analysis, aiming to increase the learning algorithms' performance. A key challenge in this field is achieving accurate attribute selection when handling both numerical and nominal attributes. To address this problem, we demonstrate a hybrid intuitionistic fuzzy (IF) similarity-based approach that flexibly handles mixed types of data for more precise attribute selection. The study shows an IF granular structure to manage noise in heterogeneous data and enhances the concepts of IF rough entropy, joint entropy, and conditional entropy to provide a comprehensive framework to deal with uncertainty. Moreover, IF rough mutual information is implemented to extract both uncertainty and the association between conditional attributes and the decision class, forming the basis of a novel attribute selection approach. The proposed algorithm contains intuitionistic fuzzification, evaluation of fuzzy mutual information to compute the significance of features, and recursive selection of the most important attributes, thus effectively reducing dimensionality. We set up a theoretical foundation using IF sets, IF information system, IF relation theory, and hybrid similarity relations, which ensures the robustness of the approach. The method is rigorously evaluated on seven benchmark datasets, showing superior performance on various metrics, including accuracy, sensitivity, and specificity, when compared with existing attribute selection approaches. Results demonstrate increased prediction of phospholipidosis-positive molecules with a sensitivity of 89.56%, a specificity of 92.63%, an accuracy of 95.98%, an AUC of 0.968, and an MCC of 0.908, which represents the strong class differentiation ability of the model. These findings underscore the effectiveness of the hybrid IF similarity-based attribute selection approach, which makes it a valuable tool for managing high-dimensional datasets and advancing the field of attribute selection.

Received: 14 May 2025 | Revised: 30 October 2025 | Accepted: 21 November 2025

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available at https://archive.ics.uci.edu/.

Author Contribution Statement

Aneesh Kumar Mishra: Conceptualization, Methodology, Validation, Data curation, Writing – original draft, Writing – review & editing. Neelesh Kumar Jain: Software, Formal analysis, Supervision, Project administration. Ravindra Kumar Singh Jain: Investigation, Resources, Visualization.