M. Agulló Antolín, E. Del Barrio Tellado
Classification rules can be severely affected by the presence of disturbing observations in the training sample. This can lead to the choice of unnecessarily complex rules. Simpler effective classification rules could be achieved if we relax the goal to fitting a good rule for a fraction of the data.
In this talk we introduce a new method based on trimming and penalization methods to produce classification rules with guaranteed performance on a significant fraction of the data. In particular, we provide an automatic way of determining the right trimming proportion and oracle bounds for the generalization error of the corresponding values.
Palabras clave: Classification, trimming, oracle inequality
Programado
L05.1 Grupo de Análisis Multivariante y Clasificación I
5 de septiembre de 2016 11:30
0.02 - Aula de proyectos 1
