Research On Clustering Algorithm Based On Particle Swarm Optimization

Data mining is defined as the process which extracts a great deal of information andknowledge from large amounts of data, these data are potential and valuable, unknown inadvance, incomplete, fuzzy, random and noisy. Clustering analysis is one of very importantresearch field, it has very important applications in such aspects as biology, medicine, businessand the Web document, is one of the hot spot of research.In such an era of data explosion, the more effective and quick approach is needed, andswarm intelligence algorithm has injected new vitality to the traditional clustering algorithm,and achieved a satisfactory clustering results. Therefore, the research of particle swarmoptimization algorithm in clustering has more realistic meaning and theory meaning. This papergives a detailed analysis on the clustering, introduces the current research situation of particleswarm optimization, gives the detailed parameter analysis and the basic ideas of particle swarmclustering, carries on the thorough analysis about the existing particle swarm clusteringalgorithm, and then proposes two kinds of improved algorithms according to the problem of theinitialization randomly. In general, this paper mainly completed the following work:⑴Particle swarm optimization algorithm initializes every particle randomly, different initialvalues have often different clustering results, the clustering result is strongly influenced by theinitial value, easy to fall into local extremum, clustering result is unstable, clustering accuracyis poor. In order to solve this problem, this paper improves the particle swarm optimizationalgorithm, the high quality of cluster center, which is obtained by affinity propagation, isregarded as an initial value of one of all particles, the other particles are initialized randomly,and then use the basic particle swarm clustering, the improved algorithm adopts the fitnessfunction of the sum of squares error and the linear differential decreasing inertia weight, theimproved method effectively avoid the defect that particle swarm algorithm is greatly influenced by the initial value. Finally experiment illustrates the proposed algorithm has higherclustering quality from three angles of the stability of clustering results, accuracy andconvergence speed.⑵Entropy-based fuzzy clustering algorithm can determine the number of clusters andcluster center, and the clustering algorithm that needs the initial value just requires theclustering center, the particles in the particle swarm algorithm need such an initial value.Therefore, this paper improves the particle swarm clustering algorithm. The cluster center,which is obtained by entropy-based fuzzy clustering, is regarded as an initial value of one of allparticles, the other particles are initialized randomly. The improved algorithm adopts the fitnessfunction of the sum of squares error and the linear differential decreasing inertia weight.Experiments show that the proposed algorithm has a certain degree of improvement inclustering stability, accuracy and convergence speed.