基于邻域模型的K-means初始聚类中心选择算法  

Initial Cluster Centers Choice Algorithm for K-means Based on Neighborhood Model

作　　者：曹付元[1,2] 梁吉业[1,2] 姜广[1,2] 

机构地区：[1]计算智能与中文信息处理省部共建教育部重点实验室,太原030006 [2]山西大学计算机与信息技术学院,太原030006

出　　处：《计算机科学》2008年第11期181-184,共4页Computer Science

基　　金：国家863计划项目(2007AA01Z165);国家自然科学基金(70471003;60773133);高等学校博士学科点专项科研基金(20050108604);教育部科学技术研究重点项目(206017);山西省重点实验室开放基金(200603023);山西省高校科技开发项目(2007103);太原市科技局科技兴市专项项目(07010724)

摘　　要：传统的K-means算法由于其方法简单,在模式识别和机器学习中被广泛讨论和应用。但由于K-means算法随机选择初始聚类中心,而初始聚类中心的选择对最终的聚类结果有着直接的影响,因此算法不能保证得到一个唯一的聚类结果。利用邻域模型中对象邻域的上下近似,定义了对象邻域耦合度和分离度的概念,给出了对象在初始聚类中心选择中的重要性,提出了一种初始聚类中心的选择算法。另外,分析了邻域模型中三种范数对聚类精度的影响,并和随机选择初始聚类中心、CCIA选择初始聚类中心算法进行了比较,实验结果表明,该算法是有效的。The traditional K-means algorithm considered as a simple method has been widely discussed and applied in pattern recognition and machine learning. However,K-means algorithm can not guarantee unique clustering result because initial cluster centers are chosen randomly, moreover, choosing initial cluster centers is extremely important as it has a direct impact on the formation of final clusters. In this paper, concepts of coupling and division are defined by using low approximation and upper approximation of object neighborhood,and importance of objects in the procedure of choosing cluster centers is also given, initial cluster centers choice algorithm for K-means based on neighborhood model is proposed. Compared with choosing initial cluster centers randomly and CCIA algorithms, cluster accuracy affected by three kinds of norm in neighborhood model is analyzed. The experimental results show that the algorithm is effective.

关 键 词：邻域模型 初始聚类中心 K-MEANS聚类 粗糙集 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论] TP301.6[自动化与计算机技术—计算机科学与技术]