Research Based On Improved Particle Swarm Optimization Algorithm C-means Clustering Algorithm

Clustering is an ancient problem and it accompanies with the human society's emergence and development and goes deeper. Mankind's wanting to know world has to distinguish different things and know similarity between things, but every earliest stage of concept formation asks for help from clustering analysis of things. Therefore research on clustering analysis has not only important theoretical significance, but also important engineering application value and humanities value. Traditional c-means clustering algorithm which can process big data set effectively is simple and run rapidly without consuming memory overly. Although c-means is one of the most common algorithms, it has two main disadvantages: its sensitivity to initial partition matrix and its being prone to be trapped in local minimal.This thesis presents an improved algorithm called c-means based on modified particle swarm optimization algorithm to overcome the traditional c-means' shortcomings. Firstly, a novel particle swarm optimization algorithm is studied. The algorithm which incorporates accelerated factor and inertia weight and increases the inertia weight linearly in the beginning and reduce it in a straight line at last, which can be proved convergent. Secondly, the improved particle swarm is applied to c-means algorithm to overcome its shortcomings. At last, computer simulation is implemented and experimental results prove its validity.