Different variants of model-based co-clustering of continuous data
Résumé
Coclustering becomes a fundamental problem that has numerous applications in different fields and particularly in bioinformatics and text mining. This kind of methods aims to obtain homogeneous blocks by clustering simultaneously a set of objets and a set of variables. To tackle the co-clustering when the objects are described by quantitative variables an adapted mixture model is proposed. Different variants are defined and, to focus on the clustering context, the maximum likelihood and the classification maximum likelihood approaches are considered. Setting these variants in this last approach, derived dissimilarities are developed, their intepretation is discussed and their impact on the quality of clustering is studied.