Abstract
Recently, dynamic clustering has attracted significant attention and has been considered as a challenging task in unsupervised classification. However, most existing approaches assume that all classification parameters are certain. Unfortunately, the reality is connected to uncertainty by nature. To solve these problems, we propose in this paper new dynamic clustering approaches, based on the well known K-modes method, under uncertainty for handling both increasing and decreasing of the clusters' number where uncertain categorical attribute values are represented and managed through the Transferable Belief Model (TBM) concepts. By using the cluster cohesion and separation concepts, our main objective is to update the clusters' partition without performing the reclustering from scratch. The experiments on known benchmark data sets, show that our dynamic methods outperform the static version.