Statistical Clustering Analysis: An Introduction - Clustering Challenges in Biological Network - page 122

Fig. 5.9. Two-dimensional dataset with one non-convex cluster.

Fig. 5.10. Different determinations of the number of clusters with different resolution levels: (a) 3
clusters with high resolution; (b) 2 clusters with low resolution.

constitute a single cluster only if this cluster satisfies our definition of cluster. Oth-
erwise, we should choose other resolution such that each cluster found satisfies the
definition, as in Fig. 5.10(a).
Scale-based method is a method which gives the numbers of clusters under
different resolution levels. The winning number of clusters is the number survives
in the largest range of resolution levels.
Scale-based method works as follows:
(0) Standardize the dataset such that each dimension has mean 0 and standard
deviation 1, denoted as X Std .Let t =0, scale parameter λ t starts from a small