Lectures and seminars SFOepi Seminar Series in Biostatistics: Roberto Rocci, Sapienza University of Rome
Title: Model based clustering in functional subspaces
Please note that this event is not filmed nor streamed – only participation in person.
Abstract
A new model-based method is proposed to simultaneously perform clustering and dimension reduction of functional data (Ramsey and Silverman, 2005). The idea is to assume that the observed functional data are distributed as a finite mixture (McLachlan G. and Peel D., 2000) of Gaussian processes. Differences among the components, in terms of means and covariances, can be represented in a functional subspace of reduced dimension. Inference is drawn conditionally on the points at which the curves are evaluated (James and Sugar, 2003) following a penalised maximum likelihood approach. The penalty is introduced to take into account for the functional nature of the data by obtaining smooth estimates of the centroids. An EM-type algorithm to compute the estimates is presented. The calibration of the penalty is performed during the iterations of the algorithm according to the idea of Gattone and Rocci (2012). The effectiveness of the proposal is tested by applications on real and simulated data.
Joint work with Stefano Antonio Gattone, University G. d’Annunzio
References
- James G. and Sugar C. (2003). Clustering for sparsely sampled functional data. Journal of the American Statistical Association, 98(462):397–408
- Gattone S.A. and Rocci R. (2012). Clustering curves on a reduced subspace. Journal of Computational and Graphical Statistics, 21:361–379
- Ramsay J. and Silvermann B. (2005). Functional Data Analysis. Springer-Verlag, New-York
- McLachlan G. and Peel D. (2000). Finite Mixture Models. Wiley, New York.
Registration: No registration is needed
If you have any questions, please contact Marie Jansson at marie.jansson@ki.se