Research On Clustering Ensemble Optimization Method

As we all know, different clustering algorithms or the same algorithm configured different parameter values may produce different clustering results. Since clustering can't solve the problem of generalization, the concept of integration is introduced. Integration technology can significantly improve the generalization ability of learning systems, and then the concept of clustering ensemble comes into existence. This paper, comprehensively analyzes and explores cluster ensemble, integrates the cluster members by regarding the evolutionary algoritim as a consistency function,and puts forward the improvement strategy; combing co-evolution with the advantages of particle swarm optimization algorithm and genetic algorithm, this paper proposes a new integrated algorithm. Experiments on the UCI data sets verify the effectiveness of the ensemble algorithm.First of all, base clustering methods used in this paper are described in detail, which include K-means,MSF and FCM method. Combing with genetic algorithms,this paper proposes a clustering integrated method based on genetic algorithm, makes a contrast of integrated performances when cluster labels and clustering center acts as chromosome respectively, and puts forward the improving strategy to achieve a better clustering performance.Then, on the analysis of the difference between particle swarm optimization (PSO) algorithm and genetic algorithm (GA),since GA uses chromosome to share information,while PSO uses the particle location, the optimization problem and the convergence speed will be greatly different because of the different information sharing mechanisms. So ensemble algorithm based on PSO is proposed, and algorithm performance is verified through experiment.At last,the co-evolution idea is described. Making use of the respective advantages of G A and PSO algorithm, this paper advances co-evolutionary clustering ensemble algorithm based on particle swarm optimization and genetic algorithm (CEGPCE), which results in the optimal search in the global scope to improve the convergence speed and local search capability. The co-evolutionary algorithm based on PSO and GA is applied to the integration as a consistency function to achieve better performance integration.