基于质心自适应选取的密度万有引力聚类算法

Density universal gravity clustering algorithm based on centroid adaptive selection

为解决典型的K-means算法中存在的k值、初始簇中心难以确定等问题,提出一种基于质心自适应选取的密度万有引力聚类算法CASG-means算法。通过质心自适应选取策略对初始簇中心进行选择,将数据集中的点划分为离群点和非离群点,将非离群点按照改进密度万有引力吸引的方式进行分配,删除空簇,自适应得到k个簇和离群点。用该算法解决K-means算法中的参数难以确定问题,将仿真结果与其它算法进行比较,验证了该算法的迭代次数、耗时和自适应聚类效果优于其它改进算法。

To solve the problems of k value and difficulty in determining the initial cluster center in the typical K-means algorithm,a density universal gravitational clustering algorithm CASG-means algorithm based on centroid adaptive selection was proposed.The initial cluster center was selected through the centroid adaptive selection strategy,and the points in the data set were divided into outliers and non-outliers.The non-outliers were allocated according to the improved density of gravity attraction,and the empty clusters were deleted.k clusters and outliers were adaptively obtained.The algorithm was used to solve the problem of parameter uncertainty in K-means algorithm,and the simulation results were compared with other algorithms.It is verified that the number of iterations,time consuming and adaptive clustering effect of the algorithm are better than that of other improved algorithms.