A Synthetic Approach to Multivariate Normal Clustering

Two methods have been suggested for grouping together observations originated from the same multivariate normal distributions, both based on a maximum-likelihood (ML) estimation, but leading to different conclusions. In this paper we compare the use of Day’s approach and that of Scott and Symons in clustering procedures from the theoretical and computational point of view. Based on this comparison, we suggest an approach unifying those approaches. The workability of the approach will be verified by numerical experiments.