Data Clustering using an Advanced PSO Variant

This paper proposes an advanced PSO variant using Subtractive Clustering methodology for data clustering. The implementation of this algorithm will be used to provide fast, efficient and appropriate solution for any complex clustering problem. This algorithm addresses the basic challenges faced with the existing PSO based clustering techniques i.e. preknowledge of initial cluster centers, dead unit problem, premature convergence to local optima, stagnation problem, etc. The proposed algorithm proved that the use of Subtractive Clustering methodology at the start of any PSO approach can improve the clustering process by suggesting good initial cluster centers and number of clusters in advance and then fasten the further clustering with the use of adaptive inertia weight factor and boundary restriction strategy. The performance of proposed algorithm is tested against well know clustering techniques over three datasets, where the results showed a better or comparable performance with respect to accuracy of clustering and convergence rate.