Feature Selection Algorithm Based on Fuzzy Mutual Information and Fuzzy Monotone

Traditional feature selection algorithms based on information theory often exhibit inherent limitations in directly addressing numerical features. To overcome this challenge, fuzzy mutual information has emerged as a widely adopted approach for handling numerical data. However, existing feature selection methods utilizing fuzzy mutual information encounter significant challenges in effectively modeling nonlinear relationships between features during redundancy analysis. In this paper, we proposed a novel feature selection algorithm that incorporates the fuzzy monotonic correlation analysis method to address these shortcomings. By leveraging the strengths of this method, the proposed algorithm enhances its capacity to capture and analyze nonlinear feature relationships, surpassing the performance of existing approaches. The algorithm is designed using a heuristic strategy and validated by an experimental classification study, which demonstrates that the algorithm has higher classification accuracy with fewer input features than several state-of-the-art algorithms.