一种基于网格距离的融合式聚类算法  

Agglomerative Clustering Algorithm Based on Grid Distance

作　　者：凌萍[1,2] 周春光[1] 王喆[1] 

机构地区：[1]吉林大学计算机科学与技术学院教育部符号计算与知识工程重点实验室,长春130012 [2]徐州师范大学计算机科学与技术学院,徐州221116

出　　处：《计算机科学》2008年第11期160-163,231,共5页Computer Science

基　　金：国家自然科学基金重点项目(60433020,60673099,60773095);国家高技术研究发展计划(863计划)(课题编号:2007AA04Z114);985工程:“计算与软件科学科技创新平台”项目;教育部“符号计算与知识工程”重点实验室资助

摘　　要：提出了一种基于网格距离的融合式聚类算法(Agglomerative Clustering algorithm based on Grid Distance,ACGD)。为规模不同的数据集分别设计了初始球状网格和初始矩形网格,并以此作为合并过程的起点。基于随机映射思想设计了网格之间的距离定义并以此完成聚类任务。ACGD的参数以自适应学习策略确定。真实数据集上的实验表明,ACGD具有良好聚类效果,具有比同类算法更高的效率和算法鲁棒性。Proposeed an agglomerative clustering algorithm based on grid distance,named as ACGD. ACGD starts initial aggregation from data grids instead of data points. For large-sized dataset and smalbsized dataset, sphere-shaped grid and rectangular-shaped initial grid were designed respectively. The aggregating process was conducted based on a novel grid distance definition, which is developed according to random projection idea. ACGD is equipped with the selftuning parameterization strategies. Experimental evidence of real datasets demonstrates the fine clustering performance of ACGD, and its advantage in efficiency and robustness over its peers.

关 键 词：网格距离 融合聚类 球状初始网格 初始矩形网格 数据粒单元 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] O241.3[自动化与计算机技术—计算机科学与技术]