粗糙邻域系统监督下的自适应密度聚类  

Adaptive Density Clustering Though Supervised Rough Neighborhood System

作　　者：李文海[1] 冯玉才[2] 马晓鸣[1] 吕泽华[2] 

机构地区：[1]武汉大学经济与管理学院,湖北武汉430072 [2]华中科技大学计算机学院,湖北武汉430074

出　　处：《小型微型计算机系统》2008年第5期841-847,共7页Journal of Chinese Computer Systems

基　　金：国家“八六三”高技术研究发展计划项目(2004AA4Z30202005AA4Z3030)资助

摘　　要：依据OPTICS可视化密度模型计算球形分布对象的密度扩张序列,指数缩减自适应水平阈值以获取聚类数量和聚类邻域;基于粗集理论计算各聚类核的上下近似区域,通过该邻域系统实现显式的对象划分方法.依据对象聚类邻域确定聚类数量和聚类核,以及对象的粗糙近似划分,使得聚类具有密度自适应和孤点不敏感的特点,取样分析有效提高了算法效率.This paper generates the density based expanding sequence of sphere-distributing objects on the basis of OPTICS viewable density model,and obtains clustering number and clustering neighborhoods by exponentially reducing the self-adaptability threshold. It also acquires the lower and upper approximation regions of all clustering kernels based on rough set theory, and achieves the explicit object-dividing method by means of their neighborhood system. It determines clustering number and clustering core according to the clustering neighborhood of the objects and divides the objects in terms of rough set approximation,so that clustering has the traits of density with self-adaptability and acnodes with non-sensitivity,and sampling accelerates the algorithms effectively.

关 键 词：近似区域 聚类核 密度聚类 粗集理论 聚类分析 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]