A Novel Clustering Algorithm Based on Fully-Informed Particle Swarm
Ekhlas Masoudi Mansour, Abbas Ahmadi · 2019
Data Clustering partitions a set of unlabeled data into several homogenous groups. Here, a new clustering method based on fully-informed particle swarm (FIPS) named FIPS-based clustering algorithm is introduced. Fully-informed particle swarm, as a generalization of standard PSO, applies different neighborhood topologies, as such, in this study we propose three topologies to be employed in clustering algorithm: all topology, distance topology and ring topology. All and ring topologies have been already used for optimization problems, however, distance topology is proposed in this study for clustering algorithm. For the purpose of evaluation, FIPS-based clustering algorithm is implemented for different problems along with canonical particle swarm optimization clustering technique using aforementioned neighborhood topologies. According to the obtained results, FIPS-based clustering algorithm with distance topology can provide much better results than canonical PSO clustering method in majority of the studied problems.