Information entropy-assisted intuitionistic fuzzy rough feature subset selection

Abstract

Fuzzy-rough set (FRS) has been effectively implemented as a powerful pre-processing tool to cope with the issues such as vagueness, imprecision, noise and uncertainty in high-dimensional data analysis. However, FRS fails to handle the uncertainty due to identification, which is very much common in heterogeneous data. Moreover, it is always challenging to handle different issues such as vagueness, uncertainty, and noise during elimination of irrelevant and redundant features. This paper investigates an intuitionistic fuzzy rough set (IFRS)-aided feature selection method by using information entropy notion to tackle these issues simultaneously. We establish an intuitionistic fuzzy (IF) similarity relation. Then IFRS model is discussed based on this relation. Next, we present a novel IF granular structure. Based on the granular structure, we present lambda-conditional entropy for IF framework. Further, relevant mathematical theorems are proved to justify the theoretical aspect of the models. Moreover, a positive region preserved attribute selection method is proposed. A comprehensive experimental study is discussed to demonstrate the effectiveness and practical validation of the proposed method. Finally, a methodology is developed based on proposed models, which increases accuracy for the minority RNA silencing suppressors against majority class non-suppressors as evident from better sensitivity and G-means metrics.