A novel robust suppressed relative entropy fuzzy C-means clustering algorithm

Commonly used entropy-based clustering algorithms, like the relative entropy fuzzy C-means clustering algorithm (REFCM), overlook the pivotal role of the centroid in iterations. Significantly, determining the precise centroid becomes challenging for REFCM in scenes characterized by complex structures or a substantial increase in the number of noise points. Therefore, this paper proposes an improved REFCM algorithm based on the gradient ascent algorithm (S-IREFCM_IF). Its advantages are as follows: (1) In order to obtain better initial results, the isolation forest algorithm is used to process the data. (2) Based on the Gaussian function, a gradient ascending iterative centroid algorithm with adaptive bandwidth parameters is proposed to determine a more accurate centroid and adapt to more data structures. (3) The suppressed operation is added to obtain a shorter algorithm running time. Experiments on synthetic datasets, and UCI datasets show that S-IREFCM_IF can obtain a more accurate centroid and better clustering performance.