基于粒计算的k值选取及其应用  

K value of k-means algorithm based on granular computing and its application

作　　者：卞彩峰 邱建林[2] 陈燕云[3] 陆鹏程[1] 陈璐璐[1] 

机构地区：[1]南通大学电子信息学院,江苏南通226019 [2]南通大学计算机科学与技术学院,江苏南通226019 [3]南通大学工程训练中心,江苏南通226019

出　　处：《计算机工程与设计》2015年第11期3082-3086,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(NSF61272424);江苏省自然科学基金项目(BK2010277);南通市科技计划基金项目(K2010002;AL2007033)

摘　　要：为降低k值选取对聚类分析算法效果的影响,采用改进的聚类有效性函数来选取k值。通过粒计算的属性分辨能力,避免数据因某些属性值过大屏蔽其它取值较小属性对类内相似性和类间差异性的影响,综合考虑类内相似性和类间差异性。通过UCI机器学习数据库标准数据集和k均值算法对该函数进行验证,并将其应用到玉米良种选育中。实验结果表明,使用该函数可以更为有效的在玉米样本集中选育玉米良种。To reduce effects of k value selection of clustering analysis,an improvement of clustering validity function was used to select k value.Attributed clustering validity function considered not only how to avoid the effect caused by data but also the difference and similarity between classes comprehensively.By using attribute resolution,the influence on the similarity within classes and the difference between classes caused by large or small attribute value was avoided.The correctness of the improved clustering validity function was verified by some datasets of UCI database and k-means algorithm.The improved clustering validity was used in corn breeding.Experimental results show that the improved clustering validity function is correct and effective in the corn samples selection while breeding.

关 键 词：聚类分析 粒计算 K均值算法 有效性函数 玉米良种选育 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]