NBCLUST: un programme de détermination du nombre de classes pour des données symboliques

Abstract

In this report, we are interested in the determination of the number of clusters for symbolic data described by combinations of classical quantitatives, classical qualitatives, intervals, multi-valued and modals variables. We adapt the best methods of determination of the number of clusters stemmed from the study of Milligan and Cooper to the program of symbolic classification Sclust. We test the method on various artificials and reals data sets and compare the results obtained with the different distances (L1, L2, Hausdorff et De Carvalho).

Date of Award	24 Jun 2005
Original language	French
Supervisor	Andre Hardy (Supervisor), Jean Paul Rasson (Jury) & Marcel Remon (Jury)