In cluster analysis it is generally assumed that one single cluster structure is contained in a data matrix, and that this structure may be confined to a subset of the observed variables. This paper investigates a new solution that simultaneously selects the relevant variables and discovers multiple cluster structures from possibly dependent subsets of variables. The basic idea is to recast the problem as a model comparison problem in which conditional independence assumptions are introduced using multivariate regression models with correlated and non-normal error terms. A stepwise procedure for selecting a locally optimal model is also proposed. Results obtained from a Monte Carlo study are briefly described.
Detecting multiple cluster structures through model-based clustering methods / G. Soffritti; G. Galimberti. - STAMPA. - (2009), pp. 263-266. (Intervento presentato al convegno 7° Meeting of the Classification and Data Analysis Group of the Italian Statistical Society tenutosi a Catania, Italy nel September 9-11 2009).
Detecting multiple cluster structures through model-based clustering methods
SOFFRITTI, GABRIELE;GALIMBERTI, GIULIANO
2009
Abstract
In cluster analysis it is generally assumed that one single cluster structure is contained in a data matrix, and that this structure may be confined to a subset of the observed variables. This paper investigates a new solution that simultaneously selects the relevant variables and discovers multiple cluster structures from possibly dependent subsets of variables. The basic idea is to recast the problem as a model comparison problem in which conditional independence assumptions are introduced using multivariate regression models with correlated and non-normal error terms. A stepwise procedure for selecting a locally optimal model is also proposed. Results obtained from a Monte Carlo study are briefly described.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.