Recently it has been shown that for k-means, Hartigan's method has better optimization performance
than the prevalent Lloyd's method. Hartigan's
method is a general idea of optimization heuristic.
When considering moving a point to another cluster,
it measures the exact change to the objective function. In this paper we develop a Hartigan's method
for another important clustering objective: k-modes,
which is popularly used for categorical data. The pro-
posed algorithm is as efficient as the Lloyd's method
on k-modes. Moreover, we rigorously prove that Hartigan's method can further improve some local optima
achieved by Lloyd's method. Empirical evaluation
verifies this conclusion. Furthermore, when these two
methods are used independently, Hartigan's method
also achieves better optimization performance.
|Cite as: Xiang, Z. and Islam, M.Z. (2014). Hartigan's Method for K-modes Clustering and Its Advantag. In Proc. Twelfth Australasian Data Mining Conference (AusDM14) Brisbane, Australia. CRPIT, 158. Li, X., Liu, L., Ong, K.L. and Zhao, Y. Eds., ACS. 25-30 |
(local if available)