Hybrid binarisation technique for the Bi-CoPaM method

A relaxed paradigm of clustering has been proposed recently in which each data object can be assigned exclusively to one cluster, assigned simultaneously to multiple clusters, or unassigned from all clusters. This has been realised by six tunable binarisation techniques for the binarisation of consensus partition matrices (Bi-CoPaM) ensemble clustering method. These techniques can be used to generate clusters with tunable tightness levels from wide clusters, through complementary clusters and towards tight clusters. In this study, we analyse these six techniques and classify them into two classes/tracks which differ in the way in which they gradually tighten clusters. We also propose using hybrid combinations of the techniques from both classes/tracks. The results of applying these techniques over a real microarray dataset of 1000 yeast genes demonstrate that, in many cases, there are significant differences between both classes/tracks of techniques. Moreover, comparisons between both classes/tracks by hybrid combinations are able to unveil information about the distinctness of the clusters and the competitiveness between them.