Model-based clustering of functional data
1 : Laboratoire Paul Painlevé
(LPP)
-
Site web
CNRS : UMR8524, Université des Sciences et Technologies de Lille - Lille I
U.F.R. de Mathématiques 59 655 Villeneuve d'Ascq Cédex -
France
This talk presents two general procedures for clustering functional data. The first one [1] approximates the density of functional random variables using the probability densities of the principal components scores. The second one [2] assumes, conditionally to the group, a Gaussian mixture model on the coefficients of the eigen-function expansion. The originality of these works is that the essential dimension reduction is achieved by group. The parameter estimation is based on the EM algorithm. Real data application illustrate the interest of the proposed methodologies.
Key-words: Functional data, functional principal component analysis, model-based clustering, random function density, group-specific functional subspaces, EM algorithm.
References
[1] C. Bouveyron and J. Jacques. Model-based clustering of time series in group-specific functional subspaces. Advances in Data Analysis and Classification, in press, 2011.
[2] J. Jacques and C. Preda. Model-based clustering of functional data. preprint HAL n00628247, 2011
