自动获取k-means聚类参数k值的算法  

Algorithm of automatic gained parameter value k based on dynamic k-means

作　　者：田森平[1] 吴文亮[1] 

机构地区：[1]华南理工大学自动化科学与工程学院,广东广州510640

出　　处：《计算机工程与设计》2011年第1期274-276,335,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(60674034)

摘　　要：为了减轻用户负担,针对k-means算法的k值给定问题进行了研究,提出了一种k值自动获取方法。对所需要聚类的数据进行抽样,对抽样数据相互之间的间距进行计算,获得密度参数R,Z和簇划分参数r,m,h。在多次抽样数据得到的参数集中,选取一组最小的参数运用到动态k-means算法中,获得k值。在自动获取参数的基础上,通过实验对簇间属性相似数据获得了准确率较高的聚类结果。实验结果表明,该算法提高了k-means算法的自动化计算能力。In order to cause user＇s less difficulties,the method of an automatic gaining of the value k is proposed to dealwith this problem after studying k-means algorithm.First,sample data should be picked out from all data.Second,computing the distance of the data between each other to get density parameters R,Z and cluster parameters r,m,h.Finally,the value k is got by applying a set of smallest parameters from multiple sampling data sets to dynamic k-means algorithm.With experiment,the accurate parameter cluster is gained on the foundation of the automatic gained parameters,which are of the similar cluster attributes.The experimental result proves the automatic computing power of k-means algorithm is improved by the new algorithm.

关 键 词：动态 K-均值 抽样数据 自动获取 密度 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]