A new approach for distributed density based clustering on grid platform

Abstract

Many distributed data mining DDM tasks such as distributed association rules and distributed classification have been proposed and developed in the last few years. However, only a few research concerns distributed clustering for analysing large, heterogeneous and distributed datasets. This is especially true with distributed density-based clustering although the centralised versions of the technique have been widely used fin different real-world applications. In this paper, we present a new approach for distributed density-based clustering. Our approach is based on two main concepts: the extension of local models created by DBSCAN at each node of the system and the aggregation of these local models by using tree based topologies to construct global models. The preliminary evaluation shows that our approach is efficient and flexible and it is appropriate with high density dataseis and a moderate difference in dataset distributions among the sites. © Springer-Verlag Berlin Heidelberg 2007.