基于加权欧式距离的k_means算法研究  

Research Based on Euclid Distance with Weights of K_means Algorithm

作　　者：张忠林[1] 曹志宇[1] 李元韬[1] 

机构地区：[1]兰州交通大学电子与信息工程学院,甘肃兰州730070

出　　处：《郑州大学学报（工学版）》2010年第1期89-92,共4页Journal of Zhengzhou University（Engineering Science）

基　　金：兰州市企业技术攻关计划资助(2009-1-4);兰州交通大学"青蓝"人才工程基金资助(QL-05-10A)

摘　　要：传统的k_means算法将欧式距离作为最常用的距离度量方法.针对基于欧式距离计算样本点与类间相似度的不足,用"相对距离"代替"绝对距离"可以更好地反映样本的实际分布,提出一种在领域知识未知的情况下基于加权欧式距离的k_means算法.针对公共数据库UCI里的数据实验表明改进后的算法能产生质量较高的聚类结果.Euclid distance is commonly used to measure distance in the traditional k_means algorithm.The k_means algorithm based on weighted Euclid distance is researched and presented to overcome the existing problems of similarity calculation in clustering analysis based on traditional Euclid distance when we have no any domain knowledge about the data objects,the relative distance but not absolute distance is more accurately response to data distribution.Experiments on the standard database UCI show that the proposed method can produce a high accuracy clustering result.

关 键 词：k_means算法 聚类 加权 变异系数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]