作 者: ;
机构地区: 暨南大学
出 处: 《数理统计与管理》 1996年第5期32-34,共3页
摘 要: 二维有序样本进行聚类必须满足两个要求:(1)类内各单元的相似性和类间的差异性;(2)各单元在位置上的有序性和类内的连通性。根据这些要求,将各单元观测指标间的距离矩阵作为聚类的指示矩将各单元之间的区位联系矩阵作为聚类的约束矩阵,在约束矩阵给出的约束条件之下,以类间单元指标的最大距离作为类间相似性指标,在指示矩阵中通过逐步聚并而将全部单元合并归类,即可得出满足要求的样本分类。 There are two requirements in cluster analysis for two-dimensional orderly sampies, (1) the homogeneity within the same ciuster and the heterogeneity between ciusters, (2) the orderliness of area units and conncetivity within the same cluster. On the basis of these requirements, the distance matrix is chosen as the indication matrix and the location linkage matrix is considered as the restriction matrix. By using the farthest neighbor mothed of hierarchical cluster analysis under these restrictions, the desired sample classification is obtained.
领 域: []