K-Means算法的研究与改进  

Research and Improvement of K-Means Algorithm

作　　者：周爱武[1] 陈宝楼[1] 王琰[1] 

机构地区：[1]安徽大学计算机科学与技术学院,安徽合肥230039

出　　处：《计算机技术与发展》2012年第10期101-104,共4页Computer Technology and Development

基　　金：安徽省教育科研重点计划项目(KJ2009A57)

摘　　要：K-Means算法是一种基于划分方法的经典聚类算法,已经在很多领域得到广泛的应用。虽然该算法有很多优点,但其也存在自身的局限性,比如需要用户输入聚类簇个数,初始聚类中心是随机性选择的,算法容易陷入局部最优解,对孤立点比较敏感等。文中首先应用统计学中的标准分数对样本进行孤立点分析,然后提出一种新的初始聚类中心确定策略。对改进的算法和原算法分别做实验进行比较,实验结果表明,改进的算法在准确率、收敛速度和稳定性方面都有很大的提高。K-Means algorithm is a classic clustering algorithm based on the classification method has been widely applied in many fields. Although the algorithm has many advantages,there are also their own limitations,such as user input the number of clusters,initial cluster centers is random selection,the algorithm is easy to fall into local optimal solution is more sensitive to outlier and so on. It firstly analyses sample outlier by statistics standard scores,and then puts forward a new strategy to determine the initial clustering centers. Improved algorithm and the original algorithm were doing experiments to compare,the experimental results show that the improved algorithm＇s accuracyrate,convergence speed and stability are improved greatly.

关 键 词：K—Means算法 孤立点 初始聚类中心 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]