Etude et applications de l'algorithme de classification recouvrante OKM (Overlapping K-Means)

Abstract

Overlapping clustering is a field in huge growth. This paper deals with this problem from the angle of OKM algorithm, based on the structure of the hard clustering method called "k-means". The key algorithm of this paper, OKM, will be commented on the basis of applications on artificial data, real data known by previous hard clustering and unknown real data. Comments will be based on performance and allocations' coherence in comparison with other selected algorithms : - SODAS's pyramidal clustering function, limited forerunner of overlapping clustering, - hard clustering method k-means on which OKM is based, - WOKM method, extension of OKM with the introduction of local weighting in clusters. Beforehand, all these clustering methods will be presented in the first theoretical part of this paper and then, in the second more practical part, will be applicated to different data sets.

Date of Award	24 Jun 2011
Original language	French
Supervisor	Andre Hardy (Supervisor), Renaud Lambiotte (Jury), Marcel Remon (Jury) & Anne Lemaitre (Jury)