Distributed K-means Clustering Using Topological Relationships

Cluster analysis has been widely studied due to its importance and several methods have been developed for this purpose.However, these methods are designed to process on centralized data.In this paper, we present an asynchronous approach based on topological relationship.The proposed approach unfolds on three steps: First, each site searches for clusters (models) of its local data.Secondly, a central site proceeds to the analysis and search for the partition of the whole (the global model).Finally, we proceed with the search for the right number of groups of the global model.We note that each local data has their own number of clusters and it can be different from the number of clusters in the entire data.The experiments have clearly demonstrated the effectiveness of the proposed approach to find the partition closest to that obtained on the data set from its subsets.