Model-based methods to identify multiple cluster structures in a data set

Model-based clustering exploits finite mixture models for detecting group in a data set. It provides a sound statistical framework which can address some important issues, such as the selection of the appropriate number of groups. As in many other cluster analysis techniques, also in model-based clustering it is usually assumed that only one relevant partition of the units exists in the data and that this unique partition is defined with respect to the whole set of observed variables. This assumption, however, may be unrealistic, especially when there is a weak dependence among some of the observed variables. In many real situations, data can be characterized by the presence of multiple cluster structures: different (independent) partitions of the units can be obtained according to different subsets of the observed variables. In these situations, units belonging to the same cluster within a given partition can be assigned to different clusters according to the other partitions. In thi...

• Type: Journal
• Year: 2007
• Where: CSDA
• Authors: Giuliano Galimberti, Gabriele Soffritti
• Keywords: Cluster Structure, CSDA 2007, Mixture Model, Observed Variables