Research On K-medoids Clustering Algorithm Based On Adaptive Particle Swarm Optimization Algorithm

With the rapid development of information technology, a great deal of data has beenaccumulated in various fields and various organizational units, many valuable informationhidden in mass of data, and people desperately want to analyze them in order to extract usefulinformation and knowledge. Thus, data mining arises and it becomes a new interdisciplinaryfield and a forward line in the data decision area. Cluster analysis is an important part of datamining technology, and its goal is to collect data based on similarity to classify, which iswidely used in pattern recognition, data compression, image processing, etc. Particle swarmalgorithm is a kind of optimization algorithm based on iteration, which finds the optimalsolution by iteration. Particle swarm optimization has been used widely for its advantages ofsimple implementation, high accuracy and fast convergence.This paper has studied the clustering algorithm based on adaptive particle swarmoptimization, and the main job includes:(1) It analyzes the advantages and disadvantages of the k-medoids clustering algorithmand the particle swarm algorithm; then uses the improved particle swarm algorithm tooptimize k-medoids clustering algorithm, by the dynamic adjustment of inertia weight factorand learning factor, and by adding the particle flight time factor to update simultaneously thespeed and position of particles, so that it is better to solve the problems of prematureconvergence and slow convergence. The experimental results show that the improvedalgorithm has better accuracy rate and the performance is relatively stable.(2) It analyzes the basic principles of the simulated annealing algorithm and proposed ak-medoids clustering algorithm based on simulated annealing and particle swarm optimization.The algorithm combines the rapid optimization ability of the particle swarm optimizationalgorithm and the probability of jumping property of simulated annealing, and improves theability of the algorithm to get rid of the local extremum. The experiments show that thealgorithm improves the convergence speed and accuracy of the algorithm.