The notion of statistical depth have recently been extended to the case of multivariate functional data. Its definition involves the choice of proper weights, averaging the univariate functional depths of each component. The choice of weights is crucial and must be carefully done according to the problem at hand. We describe a procedure that, starting from data, allows to compute a set of weights which are suitable for classification based on depths. These weights incorporate information on distances between covariance operators of the sub-populations.We show the validity of our strategy through a case study in which we perform supervised classification on ECG traces referring to both physiological and pathological subjects.
Depth measures for multivariate functional data with data-driven weights / R. Biasi, F. Ieva, A.M. Paganoni, N. Tarabelloni - In: Proceedings of the XLVII Scientific Meeting of the Italian Statistical Society / [a cura di] S. Cabras, T. Di Battista, W. Racugno. - [s.l] : CUEC, 2014 Jun 11. - ISBN 978-88-8467-874-4. (( Intervento presentato al 47. convegno Scientific Meeting of the Italian Statistical Society tenutosi a Cagliari nel 2014.
Depth measures for multivariate functional data with data-driven weights
F. IevaSecondo
;
2014
Abstract
The notion of statistical depth have recently been extended to the case of multivariate functional data. Its definition involves the choice of proper weights, averaging the univariate functional depths of each component. The choice of weights is crucial and must be carefully done according to the problem at hand. We describe a procedure that, starting from data, allows to compute a set of weights which are suitable for classification based on depths. These weights incorporate information on distances between covariance operators of the sub-populations.We show the validity of our strategy through a case study in which we perform supervised classification on ECG traces referring to both physiological and pathological subjects.File | Dimensione | Formato | |
---|---|---|---|
2834.pdf
accesso aperto
Tipologia:
Publisher's version/PDF
Dimensione
147.06 kB
Formato
Adobe PDF
|
147.06 kB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.