The classic k-means algorithm and its variations are sensitive to the choice of starting points and always get stuck at local optimal values. In this paper, we have presented a self-acting initial seed selection algorithm for kmeans clustering which estimates the density information of input points based on the theory of convex-hulls. To reach into the core of actual clusters, we successively exploit the convex-hull vertices of given input set to construct new intermediate cluster centres. We also introduce a cluster merging technique which amalgamates the similar clusters to avoid getting stuck at local optimal values. Results of numerical experiments on synthetic and benchmark (iris and ruspini) datasets demonstrate that proposed algorithm is more efficient in terms of number of true cluster, purity and normalized information gain than the classic k-means algorithm. Thus, the feasibility of our algorithm in two dimensional space was validated. © 2011 Springer-Verlag.
CITATION STYLE
Shahnewaz, S. M., Rahman, M. A., & Mahmud, H. (2011). A self acting initial seed selection algorithm for k-means clustering based on convex-hull. In Communications in Computer and Information Science (Vol. 252 CCIS, pp. 641–650). https://doi.org/10.1007/978-3-642-25453-6_54
Mendeley helps you to discover research relevant for your work.