【摘要】 密度聚类是数据挖掘和机器学习中最常用的分析方法之一,无须预先指定聚类数目就能够发现非球形聚类簇,但存在无法识别不同密度的相邻聚类簇等问题。采用逆近邻和影响空间的思想,提出一种密度聚类分析算法。利用欧氏距离计算数据对象的K近邻与逆近邻,依据逆近邻识别其核心对象,并确定其核心对象的影响空间;利用逆近邻和影响空间,重新定义密度聚类簇扩展条件,并通过广度优先遍历搜索核心对象的影响空间,形成密度聚类簇,有效解决了无法区分不同密度相邻聚类簇等不足,提高了密度聚类分析效果和效率。基于UCI和人工数据集实验验证了该算法的有效性。更多还原

【Abstract】 Density clustering is one of the most commonly used analysis methods in data mining and machine learning, and has the advantages of unprepared number of cluster in advance and finding non-spherical cluster. However, there are deficiencies such as the inability to identify adjacent clusters of different densities. This paper proposes a density clustering analysis method based on the idea of the reverse K-nearest neighborhood and the K-influence space. The Euclidean distance was used to calculate the K-nearest neighborhood and the reverse K-nearest neighborhood of the data object, the core object was identified according to the reverse K-nearest neighborhood and the K-influence space of the core object was determined. The density clustering cluster expansion condition was redefined by using the reverse K-nearest neighborhood and the K-influence space, and the K-influence space of the core object was searched by breadth-first traversal to form a density clustering cluster, so that adjacent clusters with different density were effectively distinguished, and the effectiveness and efficiency of density clustering analysis were furtherly improved. The experimental results validate the effectiveness of the cluster analysis method in UCI and artificial data sets.更多还原