Fuzzy C-Means Clustering via Slime Mold and the Fisher Score

Abstract

Fuzzy C-means (FCM) clustering has the virtue of simple structure and easy implementation; however, it relies on the initial cluster centers and is sensitive to noise. To overcome these problems, this paper presents a novel FCM clustering method with slime mold and a Fisher score. First, logistics chaotic mapping is introduced to initialize the slime mold population and increase the population diversity. Modifying the convergence factor of the slime mold enhances the convergence speed and accuracy of the slime mold algorithm (SMA). Second, an adaptive weight is introduced into the SMA to promote the transition between exploration and development. Then, this optimal solution for SMA initializes the cluster center of FCM to avoid initialization sensitivity. Third, when considering the influence of feature differentiation degrees on the samples, the feature evaluation criteria of the Fisher score is constructed and then the importance of the feature is ranked to identify noise. The square root error criterion selects the most effective features to improve the clustering effect. Finally, by constructing uncertainty relations and introducing information entropy, the objective function of FCM is constructed to effectively solve the issue of FCM being sensitive to noise. The experimental results on 13 benchmark test functions for optimization, and 25 datasets for clustering show that the proposed algorithm outperforms other compared algorithms in terms of several evaluation metrics.