Algorithms for Model-based Block Gaussian Clustering

When the data consists of a set of objects described by a set of continuous variables, the clustering can concern the sets of objects (rows), variables (columns) or the both sets simultaneously. Considering the last type of clustering, we propose a new mixture model and develop an adapted Generalized EM (GEM) algorithm as part of the maximum likelihood, and a Classification GEM (CGEM) version as part of the classification maximum likelihood approach. The different steps of these new algorithms are presented showing the interest in data mining context. Some illustrative synthetic data allow us to evaluate their performances in comparison with EM and Classification EM (CEM) applied on the sets of objects given a partition of variables, and EM and CEM applied only on the set of objects.