Two Methods of Non-hierarchical Clustering

Abstract

As mentioned in the Preface, the development provided in this book is dominated by the potential of applying ascendant agglomerative hierarchical clustering to all types of data. Nonetheless, the specific methodology devoted to non-hierarchical clustering is also very important. In these conditions, we shall describe in this chapter two mutually very different methods of non-hierarchical clustering. The first one, called “Central Partition” method, is due to S. Regnier (I.C.C. Bull. 4:175–191, 1965 [35], Revue Mathematiques et Sciences Humaines 22:13–29, 1983 [36], Revue Mathematiques et Sciences Humaines 22:31–44, 1983 [37]). The second method called “methode des Nuees Dynamiques” or “Dynamic cluster method” is due to E. Diday and collaborators (Revue de Statistique Appliquee XIX(2):19–33, 1971 [10], J. Comput. Inf. Sci. 2(1):61–88, 1973 [11], Recherche Operationnelle 10(6):75–1060, 1976 [16], R.A.I.R.O. Inf. Comput. Sci. 11(4):329–349, 1977 [13]). This approach corresponds to a vast generalization of the K-means method, initiated by Forgy (Biometrics, Biometric Society Meeting [17]) and Jancey (Aust. J. Bot. 14:127–130, 1966 [19]).