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

Cluster analysis as an important part of information processing technology,has been used in pattern recognition,image processing,data analysis and others fields. K-means clustering algorithm is commonly used in cluster analysis, the algorithm is simple and easy to implement However, it’s liable to trap in a local optimum and sensitive to the initial center. In order to overcome the shortcoming, this paper modifies particle swarm optimization algorithm and proposes two optimized clustering algorithms:(1)For the purpose of fixing initialization sensitive issue of the K-means algorithm, this paper proposes a hybrid clustering algorithm based on modified particle swarm optimization and k-means clustering. The new clustering algorithm uses K-means to divide particles into several categories and then choose the optimal clustering domain to produce vaccine. After that, it adopts vaccination and immune selection to improve the diversity of particles. The k-means clustering algorithm is used to improve the convergence precision when the fluctuation of the particle swarm is less than the threshold setting. Experimental results show that the algorithm accuracy is higher compared with other algorithms.(2)In order to solve the defect that K-means clustering algorithm traps into local optima easily, this paper proposes an immune particle swarm optimization clustering algorithm based on perturbation strategy. In this new algorithm, disturbed operators are introduced to improve the ergodicity of particles by changing movement directions of particles when continuous stagnation exceeded the setting threshold. As a result, it improves algorithm’s convergence effect.Experiments show that, compared with other relatively new algorithm, this algorithm has higher accuracy and better stability.