D-GridMST: Clustering Large Distributed Spatial Databases

In this paper, we will propose a distributable clustering algorithm, called Distributed-GridMST (D-GridMST), which deals with large distributed spatial databases. D-GridMST employs the notions of multi-dimensional cube to partition the data space involved and uses density criteria to extract representative points from spatial databases, based on which a global MST of representatives is constructed. Such a MST is partitioned according to users’ clustering specification and used to label data points in the respective distributed spatial database thereafter. Since only the compact information of the distributed spatial databases is transferred via network, D-GridMST is characterized by small network transferring overhead. Experimental results show that D-GridMST is effective since it is able to produce exactly the same clustering result as that produced in centralized paradigm, making D-GridMST a promising tool for clustering large distributed spatial databases.