A Novel Spectral Clustering Algorithm Based on Randomly State Changed Particle Swarm Optimization

Spectral clustering algorithm is a method of clustering which allows one piece of data to belong to two or more clusters. In this paper, a novel spectral clustering algorithm based on randomly state changed particle swarm optimization is proposed. The initial population was classified by considering global and local optimal functions, the evolutionary state of each particle is considered through the comparison of cost functions, and the evolutionary state of the particles was subdivided. The Weight mode added the previously optimal local particles and global particles. According to the rule that newer particles have greater weights, particles speed was updated to reduce the possibility of falling into a local optimal state and to expand the search range of particles. The classification accuracy of the clustering algorithm was presented. Finally, by using the UCI datasets for comparison, it was found that the algorithm proposed in this paper increase the performances by 3% to 28% for different datasets, comparing with the known clustering algorithms such as particle swarm optimization algorithm, Fuzzy C-Means algorithm and conventional Spectral Clustering.