对聚类算法普遍存在问题的解决办法  

Solutions to General Clustering Algorithmic Issues

作　　者：姜园[1] 张朝阳[1] 仇佩亮[1] 戚玉鹏[1] 

机构地区：[1]浙江大学信息与通信工程研究所,浙江杭州310027

出　　处：《电路与系统学报》2004年第3期92-99,共8页Journal of Circuits and Systems

基　　金：国家自然科学基金资助项目(60002003)

摘　　要：聚类广泛应用于统计、机器学习、模式识别、数据分析等领域并越来越受重视。本文研究了各种聚类算法共同面临的五个问题:聚类效果评估、类数目估计、数据预处理、样本间相似性测量、抗干扰性能,分析了对这些问题的有代表性的解决方法,总结并预测了未来聚类算法在这五个方面的研究方向。Clustering is widely used in several fields such as statistics, machine learning, pattern recognition and numerical analysis. Recently, more and more attention has been paid to it. In this paper, five issues commonly concerned are discussed, they are: assessment of clustering results, estimation of total number of clusters, data preparation, measures of data proximity and outlier handling. Representative solutions to these issues are surveyed, conclusions are summed up, development trend of algorithms to deal with these five issues is forecasted.

关 键 词：聚类 效果评估 类数目估计 预处理 相似性测量 抗干扰性能 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]