Parameter-wise co-clustering for high-dimensional data

In recent years, data dimensionality has increasingly become a concern, leading to many parameter and dimension reduction techniques being proposed in the literature. A parameter-wise co-clustering model, for (possibly high-dimensional) modelled via continuous random variables, is presented. The although allowing more flexibility, still maintains very high degree of parsimony interpretability achieved by traditional co-clustering. More precisely, keystone consists dramatically increasing number column-clusters while expressing each as combination limited mean-dependent variance-dependent column-clusters. stochastic expectation-maximization algorithm along with Gibbs sampler used estimation an integrated complete log-likelihood criterion model selection. Simulated real datasets are illustration comparison