Research Of K-means Algorithm And Parallelism Based On Hybrid Particle Swarm Optimization

Data mining has four main task, association analysis, cluster analysis, predictive modeling, and abnormal detection. The clustering analysis is one of most important and broad using method. High efficiency and high precision result is one of data mining goal. People do many researches for this goal. One research introduces other algorithm in data mining. These algorithms include intelligent algorithm, heuristic algorithm, neural network, fuzzy theory, and rough set theory and so on. Here introduce taboo search and particle swarm optimization in K-means cluster algorithm in the paper, which can be improved efficiency of the K-means clustering algorithm and precision of the cluster result.Taboo search is an intelligent, heuristic, and globe neighborhood search algorithm. The algorithm use part neighborhood search mechanism and taboo rule to avoid repetitive search, and use breaking taboo level to release the being tabooed fine object, so as to ensure the diversified and effectual search. Research show that taboo search can conquer evolvement algorithm easy getting into disfigurement of earliness, this algorithm can find the globe optimization, finally. Particle swarm optimization is an evolvement calculates technology. It is simple, effectual, quick convergent speed. It is fit for colony behavior research. This algorithm be got high attention by academe in recent years, but it easy get into local optimization, It consequently lead to low precision result and slow convergent speed. This paper uses taboo search and parameter control to improve on particle swarm optimization, thereby the efficiency of algorithm and precisions of result are heightened. K-means is a clustering method based on partition. It broad uses in cluster analysis at recent years. But it easy gets into local optimization and has low efficiency. Moreover, the number of clustering K is often attains according to experience or random selected. So it will effect on clustering result. To the K-means algorithm shortage, here introduce taboo search and particle swarm optimization in K-means cluster algorithm, so order to improve efficiency of the K-means clustering algorithm and precision of the clustering result. In paper discuss taboo object, selecting taboo table structure and individual coding mode, improving on inertia weight, selecting or constructing penalty function mode and expression, constructing fitness function. Experiment result of improved K-means algorithm shows that efficiency and precision results are improved. To more improve the running efficiency of algorithm, in paper discusses how to realize parallel K-means algorithm. The parallel program selects Master/Slave model. Using equivalence relation between population and sub population establish equivalence class, according to equivalence classes to part the population, then send the population to slave node, realize data parallel, finally. The master node gathers clustering results from each slave node and collects the final result. The paper evaluates the parallel algorithm from theoretical study. Theory analysis shows that parallel algorithm can obtain a good clustering result as well as serial algorithm and upgrade the efficiency.