Pattern Recognition, 4th Edition

15.12. Combination of Clusterings

Throughout the second part of the book, we discussed algorithms that produce a single clustering (or a hierarchy of clusterings) for a given data set X = {x¹,x²,…,x^N}, xⁱ ∈ R¹, i = 1,2,…,N. In this section, the situation is different. A (final) clustering is obtained based on a set of n different clusterings of X. Specifically, the aim here is two-fold: (a) the production of an appropriate set of clusterings, ∈, for the data set, X, called ensemble of clusterings, and (b) the combination of the clusterings of ∈ to produce a final clustering, called consensus clustering. The main motivation for considering such techniques is that it is expected that the resulting consensus clustering, based on ∈, will model ...