A Novel Approach of Data Clustering Using An Improved Particle Swarm Optimization Based K–Means Clustering Algorithm

In this paper, a Modified Particle Swarm Optimization (MfPSO) based K-Means algorithm is presented to cluster multidimensional data. Poor selection of cluster centers in K-Means at the initial stage may affect the clustering result and it may get stuck at local minima. To get rid of these problems, the proposed MfPSO is employed to generate the cluster centers for a dataset. The inertia weight of PSO algorithm plays a vital role to balance the global search and local search in PSO. In the proposed algorithm, the inertia weight has been modified to improve the convergence velocity and better global search capability. The MfPSO generates the cluster centers and those derived cluster centers are then applied as the initial cluster centers in the K-Means algorithm. It has been proved quantitatively that the proposed algorithm produces better result and the local minima problem has been resolved. The proposed algorithm has been compared extensively with the conventional K-Means algorithm and chaotic descending inertia weight based PSO (CDIW PSO) on four well known dataset. The superiority of the proposed algorithm is visually and quantitatively established on the basis of two standard cluster evaluation criteria, computational time, mean, standard deviation of fitness and two other statistical significance test, called ANOVA test and t-test, best fitness curves and convergence curves for different levels of clustering.