A Cluster Algorithm Based On Improved Particle Swarm Optimization And K-Means

Cluster Analysis is the important method of Data Mining,and apply to differentfields in Pattern Recognition、Data Analysis and Market Research. Particle Swarmoptimization is the Biological Optimization Algorithm,which is gradually developed inrecent years and widely concentrated and investigated because of its advantages.We firstly analyse the basic conception of PSO and the theory of existingalgorithms, propose a improved PSO──FWPSO algorithm. Then analyse the methodand catelogue of Cluster Analysis and propose a improved one──VBK-means.Finally, we combine two of them to raise the upgraded one──FWP-VBK. Including:1. Raise the updated FWPSO. It introduce inertia weight from the diffreencebetween Particle fitness and Particle evaluation to get the Convergence during theParticle Iterations. FWPSO apply Fitness Evaluation to revise the Inertia Evaluation andLearning element to dynamically change flying speed and adapt its proportion oflearning and group learning according to its Convergence conditions so as to beoutside the locally optimal solution and strengthen the function of its searching andaccuracy.2. We raise the updated VBK-means clustering algorithm base on the shortcomingof K-means clustering algorithm. We analyse relations in intra-cluster variation andinter-cluster variation and Clustering Effect to raise Balanced Variation that used to theVBK-means algorithm as the Evaluation Function so as to get the best Clusteringresults.3．The VBK-means Clustering Algorithm has low efficiency to handle massdata,but FWPSO has high capability of searching to handle mass data.We finally raisethe upgraded FWP-VBK Clustering Algorithm on the basis of FWPSO and VBK-means.As tested, the FWP-VBK Clustering Algorithm efficiently overcome the shortcoming ofK-means Clustering Algorithm and make much progress in algorithm validity andefficiency.