Clustering data stream under a belief function framework

Abstract

Clustering is a crucial task for massive data that continuously arrive and evolve over time, generated as stream. However, data may be pervaded by uncertainty and imprecision, and techniques that achieve the unsupervised learning with imperfect data sets are unable to deal with such evolving environment. On the other hand, standard methods for clustering data streams are not adapted to an uncertain framework. Hence, in this paper, we propose a method for clustering data stream in an imperfect context, particularly using belief function theory in order to handle the belonging of objects to singletons and disjunctions of clusters.