Research On K-means Optimization Clustering Algorithm

Clustering is a very important technology of the data mining. According to certain rules, the function of clustering is to devide the large data sets into groups.K-means algorithm is widely used for clustering algorithm.This paper deeply analysis and studies of K-means algorithm.K-means algorithm is easy to be achieved and high efficien- t.However,K-means has some defects——sensitive to initial value, easy to be impacted by outlier, easy to get into a local optimum. For this reason, this paper designed two kinds of improved algorithm. Main work has been done as follows:1. Aiming at sensitive to initial value ,initial value optimization method is used for clustering. First of all,a method is designed that is used to initializa center based on density, distance and neighborhood. then, K-means algorithm is improved by the method.Then, further combined the thoughts of dynamic clustering and rough clustering,a kind of K-means rough clustering algorithm is designed.Finally, experimental results show that the improved algorithm compensates for shortage of K-means algorithm in more degree, improving stability and effectiveness of clustering results.2. Aiming at easy to get into a local optimum,hybrid algorithm have been designed to realize clustering. The paper provided a systematic analysis and study of the characteristics of differential evolution algorithm and K-means.On this basis,K-means clustering algorithm based on differential evolution algorithm have been designed. The method is designed on the base of the organic combination of the two.The method gives full play to local search ability of K-means algorithm and global optimization ability of differential evolution algorithm. Experimental results show that the cluster quality can be more effectively improved.