Clustering Analysis And Application Based On Particle Swarm Optimization Algorithm

As one of the most important tool for data mining and pattern recognition, clusteringanalysis has been widespread widely and has been a hot topic. C-means clustering algorithmis the most widespread and popular in all clustering algorithms. It has not only a deepmathematical foundation, but also has been used successfully in many areas. However its vitalshortcoming is sensitive to the initial values and easy to trap into a local optimal solution,which may cause the randomness of clustering results, affecting its effect.Particle swarm optimization (PSO) is a widely applied global optimum algorithm, and itsmain merits are simplicity, easy manipulation, as well as the functions of currency andmemory. Combining PSO algorithm with C-means clustering algorithm, we will get a hybridalgorithm which has good global and local search capability. It can enhance convergencespeed of genetic algorithm (GA), PSO and FCSS. The new algorithm uses crossover andmutation factors in genetic algorithms to optimize the particle position, and to increase itscapacity of convergence to solve the problem of spherical shells data clustering.This paper is engaged in the hybrid algorithm of PSO and traditional clusteringalgorithm. Firstly, we introduce FCSS (fuzzy c spherical shells ) and use it to cluster fornon-concentric spherical shells and concentric circles experimental data sets. As shown by theresult of clustering tests, FCSS algorithm's clustering quality for non- concentric sphericalshells data sets is satisfying, but the clustering for concentric circles data sets is not effective.So we present an improved algorithm based on PSO and FCSS called PSO-FCSS clusteringalgorithm for concentric circles data sets. Compared with the results of PSO-FCSS andGA-FCSS(Fuzzy C-spherical shell cluster algorithm based on genetic algorithm and FCSS),convergence speed of PSO-FCSS algorithm is faster than GA-FCSS, however with increaseof the number of data points and classes, the performance of the PSO-FCSS algorithm isunsatisfying ; GA-FCSS algorithm has better global convergence, and its defect is theconvergence speed that is unsatisfying. Taking these factors, we present an efficient hybridspherical shell clustering algorithm called PSO-GA-FCSS, which is based on the combinationof genetic algorithm (GA), PSO and FCSS. The new algorithm uses crossover and mutationfactors in genetic algorithms to optimize the particle position, and to increase its capacity of convergence to solve the problem of spherical shells data clustering.Furthermore, this paper uses a hyprid clustering algorithm based on PSO and C-meansclustering algorithm to solve the problem of telecom customer classification. During theprocess, then this paper uses particle swarm optimization algorithm to generate initial solutionfirstly, iterates to generate new solutions. In the late of the iteration, and optimize s newindividuals by the C-means algorithm to improve the convergence rate. The new algorithmhas no degradation, which can converge in a very smooth way and get a good solution tosolve the problem of customer classification, which helps to make and adjust marketingpolicy for enterprises.