Applying a particle swarm optimizer controlled by a fuzzy logic controller in a multi-pivot means clustering algorithm

The fuzzy c-means clustering algorithm (FCM) has been successfully applied in different fields. However, previous studies showed that data points near boundaries between different clusters are easily misclustered. It becomes crucial to improving the clustering results of the boundary data points. Co...

Full description

Saved in:
Bibliographic Details
Published in:Applied soft computing Vol. 186; p. 114118
Main Authors: Xia, Xuewen, Li, Minghan, Song, Haojie, Xu, Xing, Zhang, Yinglong
Format: Journal Article
Language:English
Published: Elsevier B.V 01.01.2026
Subjects:
Center-based clustering algorithms
Data clustering
Multi-pivot strategy
Particle swarm optimization
Center-based clustering algorithms
Multi-pivot strategy
Particle swarm optimization
Data clustering
ISSN:1568-4946
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The fuzzy c-means clustering algorithm (FCM) has been successfully applied in different fields. However, previous studies showed that data points near boundaries between different clusters are easily misclustered. It becomes crucial to improving the clustering results of the boundary data points. Considering that a single center in each cluster does not contain comprehensive distribution information of the cluster, we hybrid a multi-pivots strategy with FCM, and propose a multi-pivots means algorithm (MPM). In MPM, we firstly utilize the basic FCM to perform the clustering. Then, a controversial area (CA) concept is introduced to describe the overlapping area between different clusters. Based on the CA, all data points can be divided into controversial points (CPs) and deterministic points (DPs), which are inside or outside of the CA, respectively. Moreover, a particle swarm optimization algorithm with adaptive learning weights by a multiple-input multiple-output fuzzy logic controller (MFCPSO) is used to optimize multiple pivots in each cluster. Lastly, based on the optimal multiple pivots, the CPs, which are easily misclustered, are reclustered, intending to improve their classification accuracy. The experimental results and comparisons between MPM and other 6 clustering algorithms on 10 popular datasets suggest that MPM exhibits auspicious performance on different datasets. Furthermore, the effectiveness and efficiency of the introduced strategies are also discussed based on a set of experiments. •Data points are divided into “controversial points" (CPs) and “deterministic points" (DPs) based on the proposed concept called “controversial area" (CA).•A multi-pivot strategy is introduced in each cluster.•The positions of multiple pivots in the cluster are optimized by a novel PSO variant.•Based on the optimized multiple pivots, the CPs are reclustered aiming to improve the clustering quality.
ISSN:1568-4946
DOI:10.1016/j.asoc.2025.114118