Determining the number of clusters in the Straight K-means: Experimental comparison of eight options

The problem of determining “the right number of clusters” in K-Means has attracted considerable interest, especially in the recent years. However, to the authors’ knowledge, no experimental results of their comparison have been reported so far. This paper intends to present some results of such a comparison involving eight cluster selection options that represent four different approaches. The data are generated according to a Gaussian-mixture distribution with the clusters’ spread and sizes variant. Most consistent results are shown by the silhouette width based method by Kaufman and Rousseeuw (1990) and iKMeans by Mirkin (2005).