基于测度的网格聚类算法  

A Clustering Algorithm Based on Counting Measure

作　　者：白鹭[1] 马骥[1] 

机构地区：[1]沈阳大学信息工程学院,辽宁沈阳110044

出　　处：《沈阳大学学报》2009年第4期61-64,共4页

基　　金：基于人工免疫模型机械系统早期故障预警技术研究(E051303)

摘　　要：基于测度的网格聚类方法在数据空间上定义计数测度,并以计数测度构造目标函数.通过调整划分数据空间的分辨率,使目标函数值最大,从而实现分辨率的自动确定.在此分辨率下,某些数据细节被忽略,但是数据的主要属性和关系更为明显.距离较近的数据将被聚类到属性相同的同一簇中,使簇间的数据相似性最小,簇内的数据相似性最大.算法中没有对参数值进行人为设定,可以实现提高准确性的目的.The clustering algorithm based on counting measure defines counting measure on data space. An objective function is constructed with the counting measure which is defined on the space. The value of the objective function could be maximized through adjusting the resolution ratio of dividing the data space. Thus the resolution ratio of dividing the data space could be determined automatically. At the determined resolution ratio, some details of the datum are hidden, but key attributes and relatives are more obvious. And data which are close in distance could be clustered into one cluster in which the elements are alike. Thus the data in the same cluster are most similar, and data in different clusters are most different. Any parameter in the algorithm is determined through counting, thus the accuracy of this clustering algorithm is improved.

关 键 词：聚类 算法 计数测度 分辨率 

分 类 号：TN911.1[电子电信—通信与信息系统]