Shared Near Neighbor Maximal Spanning Trees for Cluster Analysis,

摘要

Combining shared near neighbor rules with spanning trees in a graph theoretic framework is shown to provide a very powerful generalized clustering analysis tool which is both effective over a wide range of clustering problems and computationally elegant. The clustering method proposed is developed against the background of agglomerative,hierarchical,single link,complete link,minimal spanning tree and other related approaches which are abundantly represented in the literature. Many experimental results are presented,and comparisons are made with two-dimensional data to aid subjective appreciation of point data structures and the applicability potentials of the illustrated methods.