A novel algorithm for initializing clustering centers
Shuzhong Yang, Siwei Luo · 2005
It is known that many clustering algorithms which converge to one of numerous local minima through an iterative procedure are especially sensitive to initial clustering centers. In this paper we propose a novel algorithm for refining initial clustering centers. In the algorithm we define two new measurements to measure a point's local density and then produce a clustering center with local maximal density for each cluster using either of measurements. After refinement, these clustering algorithms which are sensitive to initial clustering centers will converge to a "better" local minimum more efficiently and more rapidly. Experiments demonstrate that the proposed algorithm is feasible and efficient.