Multivariate outlier detection with robust Mahalanobis distance based on Shrinkage
A collection of robust Mahalanobis distances is proposed to address the problem of detecting outliers in multivariate data. The distances are based on robust location and covariance matrix estimators using the notion of Shrinkage. The parameters needed for the shrinkage estimators defined, have been optimally estimated. By means of a comparison with other existing methods from the literature, the performance of the proposal is illustrated in simulated scenarios and with a real dataset example. The results, especially for high dimension, about the correct and false classification rates, the behavior with skewed or heavy-tailed distributions and the inexpensive computational times show the competitiveness of our proposal.
Palabras clave / Keywords: distance-based methods multivariate analysis outliers robust Mahalanobis distance shrinkage
Otros trabajos en la misma sesión
Últimas noticias
-
04/06/18
Certificados -
13/04/18
Resumen del programa y Programa detallado -
22/03/18
Descuentos en medios de trasporte para congresistas y acompañantes -
01/02/18
Ampliación del plazo de tarifa superreducida -
19/01/18
Ampliación de plazos -
15/01/18
Programación para el día 29 de mayo -
15/01/18
Conferenciantes plenarios -
12/01/18
Sede: Palacio de Congresos -
24/12/17
Sesión plenaria en memoria del Profesor Pedro Gil -
24/12/17
Corrección bases del Premio Ramiro Melendreras
