Matrix Normal Cluster-Weighted Models

Abstract Finite mixtures of regressions with fixed covariates are a commonly used model-based clustering methodology to deal regression data. However, they assume assignment independence, i.e., the allocation data points clusters is made independently distribution covariates. To take into account latter aspect, finite random covariates, also known as cluster-weighted models (CWMs), have been proposed in univariate and multivariate literature. In this paper, CWM extended matrix data, e.g., those where set variables simultaneously observed at different time or locations. Specifically, cluster-specific marginal conditional responses given assumed be normal. Maximum likelihood parameter estimates derived using an expectation-conditional maximization algorithm. Parameter recovery, classification assessment, capability Bayesian information criterion detect underlying groups investigated simulated Finally, two real applications concerning educational indicators Italian non-life insurance market presented.