一种优化初始中心的改进K-means算法  

Improved K-means algorithm on optimized initial center points

作　　者：赵京胜[1] 韩凌霄[1] 孙宇航[1] 

机构地区：[1]青岛理工大学通信与电子工程学院,青岛266520

出　　处：《青岛理工大学学报》2015年第6期99-102,共4页Journal of Qingdao University of Technology

基　　金：国家自然科学基金资助项目(61173056)

摘　　要：传统的K-means算法由于随机选取初始簇中心,造成聚类结果不稳定,容易陷入局部最优.针对这个问题,提出了一种优化初始中心的方法,即在高密度区域中以距离最远的两点作为初始的簇中心,然后再找到这两个初始中心距离和最大的点作为第3个初始中心,依此类推,直到找到k个初始中心.实验结果证明,改进的K-means算法,有较好的准确率,能够消除算法对初始中心的依赖,提高了聚类效果.Initializing cluster centers randomly,the traditional K-means algorithm is prone to local optimal and instability of clustering results.To solve the problems,this paper presents a method of optimizing the initial center.The algorithm selects two points at the furthest mutual distance in high-density region as the initial cluster centers;then this paper sets up the third initial center according to the maximum distance with kpoints found this way.The experimental results demonstrate that the improved K-means algorithm has a better accuracy,as can eliminate the dependence on the initial cluster center,and the clustering effect has been greatly improved.

关 键 词：K-MEANS算法 初始聚类中心 高密度区域 最大距离 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]