Research On Three Key Issues Of Cluster Analysis Based On Particle Swarm Optimization

In recent years,particle swarm optimization algorithm has been widely used in many fields such as pattern recognition,spam detection,data clustering,robotics,recommendation system and so on.However,the traditional particle swarm optimization algorithm still needs to be further studied in terms of cluster validation,updating rules of velocity and position,convergence performance and so on in different application areas.Therefore,aiming at three key issues about cluster validity index,clustering algorithms and the complex community detection,this paper puts forward the cluster validity index of dynamically terminating the clustering process.This paper puts an emphasis on PSO-based clustering algorithms.The primary research works of the thesis are given as follows.1.We proposed a method to cluster validation of dynamically determining the optimal cluster number based on proposed multiple cluster measures.The method utilizes the proposed cluster validity index-Ratio of Deviation of Sum-of-squares and Euclid Distance(RDSED).In this paper,the current RDSED is calculated according to the adjacent DSED value and the clustering process is terminated dynamically.Experimental results on artificial datasets and the real-world datasets show that the RDSED can effectively evaluate the clustering results and determine the optimal clustering number.2.We proposed a improved PSO-based K-means clustering Algorithm(KIPSO).Compared with the traditional particle encoding method,KIPSO adopts a reduced particle encoding method,which normalizes the data object.The cluster process is carried out by using the average distance between the data object and the cluster center.KIPSO combines the PSO algorithm and the K-means algorithm,and has strong global searching ability of PSO and the local search ability of the K-means algorithm.Experimental results on artificial and real-world datasets show that the proposed algorithm is more accurate and the performance on convergence is better.3.A novel discrete particle swarm optimization algorithm with particle diversity and mutation(DPSO-PDM)is proposed in complex network community detection,which redefines the updating rules of the particles of velocity and position.DPSO-PDM utilizes two evolutionary strategies.Experiments on GN benchmark networks and real-world networks show that the proposed algorithm can effectively detect the network community structure,and the community detection quality and the global convergence are satisfactory.The research contributions of this paper are as follows.The difference between different indexes in the process of cluster validation is measured from the aspects of separation measures and compactness measures,and the validation process is terminated dynamically.The traditional clustering algorithm based on particle swarm algorithm is optimized.In addition this paper redefines discrete complex community detection algorithm based on particle swarm optimization.The experimental results show that the proposed method and algorithms are effective and feasible.