Clustering with the Average Silhouette Width

The Average Silhouette Width (ASW) is a popular cluster validation index to estimate the number of clusters. question whether it also suitable as general objective function be optimized for finding clustering addressed. Two algorithms (the standard version OSil and fast FOSil) are proposed, they compared with existing methods in an extensive simulation study covering known unknown numbers Real data sets analysed, partly exploring use new non-Euclidean distances. ASW shown satisfy some axioms that have been proposed quality functions. prove useful sensible many cases, but weaknesses highlighted. These concern estimating clusters together other methods, which interest due popularity this task.