Gravity based spatial clustering M. Indulska and M. E. Orlowska   View Paper (PDF)   Return to Data clustering Abstract In this paper we examine recent work in the area of spatial clustering with obstacles (14, 13) and present a discussion of several identified drawbacks. We propose an algorithm, called GRAVIclust, which addresses the identified problems. The algorithm uses a heuristic to pick the initial cluster centres and utilises centre of cluster gravity calculations in order to arrive at the optimal clustering solution. We show that the proposed algorithm not only calculates better initial cluster centres than those calculated in (13) but also that with each iteration of the algorithm we get closer to the optimal clustering solution (as indicated by the converging distance function) as opposed to the randomised results offered by (13). DiSC'03 © 2003 Association for Computing Machinery