Studies Of Cluster Analysis Improving Particle Swarm Optimization (PSO)

With the development of information processing technology, computers are used indada statistics and administration, and as such data scale swells continuously. As aconsequence, how to extracting valuable data has become the focus in the study of dataprocessing. This promotes the appearance of data mining, which is currently one of themost advanced researches in database and information decision. As an important branchof data mining, cluster analysis has been widely applied in many fields, such as markettrend research, user behavior study, a variety of pattern recognition, analysis of bigdatasets, graphics and image processing, and so on.Nowadays, many researchers have been devoting themselves to cluster analysis,publishing many works and putting forward various cluster algorithms. However,cluster algorithm varies with cluster data and their application. Based on the basictheories of cluster algorithm and particle swarm algorithm in data mining and theircurrent development, this paper points out the strong and weak points of particle swarmoptimization by comparing it with the current cluster algorithm, and improves thecurrent cluster algorithm by avoiding trapping in local optimization in processinghigh-dimensional data, as follows:(1) The revolutionary theory is applied to solving cluster problems, on the basis ofimproved objective function, according to the basic particle swarm optimization.(2) Combined with clone selection mechanism, the clone selection operator can dothe whole or part search around the same particle in all directions, and can push thequick revolution of particles in cluster, so that the stability and reliability of clusteralgorithm can be improved by solving the cluster problem of high-dimensional data, andeffectively avoiding local optimization.(3) After improving the previous algorithms, this paper, in the end, makes acomparison between the improved fuzzy clustering method (FCM), particle swarmoptimization fuzzy clustering method (PSOFCM) and clone selection fuzzy clusteringmethod (CSFCM) in the way of cluster accuracy, and carries out the simulatingexpriment wiht Eclipse integrated software. The testing result shows the high stability and reliability of the improved clustering algorithm in solving high-dimensional data.