A new approach to post-hoc problem in cluster analysis

Article type: Research Article

Authors: Dronov, Sergei V.^{a; *} | Dementjeva, Elena A.^b

Affiliations: [a] Department of Mathematics, Altai State University, Barnaul, Russia | [b] Orthopedic Stomatology Department, Altai State Medical University, Barnaul, Russia

Correspondence: [*] Corresponding author. E-mail: dsv@math.asu.ru.

Abstract: Suppose that a cluster partition of some finite set is constructed using several given characteristics of the set elements. Our aim is to arrange these characteristics in descending order with a respect to the strength of their influence on the cluster structure. To manage this, we propose a numerical coefficient presenting the degree of fixedset of characteristics impact on clusters. When we have arranged the characteristics with a respect to the proposed coefficient, we get proper solution for the problem. Our approach also allows studying connections of the characteristics in view from the point of the magnitude of changing in the cluster structure.

Keywords: Clusters, set partitioning, post-hoc analysis, metric on final sets

DOI: 10.3233/MAS-2011-0201

Journal: Model Assisted Statistics and Applications, vol. 7, no. 1, pp. 49-55, 2012