Robust statistical boosting with quantile-based loss functions

• Jan Speller ^[1] ; Christian Staerk ; Andreas Mayr
  1. [1] University of Bonn

     University of Bonn

     Kreisfreie Stadt Bonn, Alemania

     
• Localización: Proceedings of the 35th International Workshop on Statistical Modelling : July 20-24, 2020 Bilbao, Basque Country, Spain / Itziar Irigoien Garbizu (ed. lit.), Dae-Jin Lee (ed. lit.), Joaquín Martínez Minaya (ed. lit.), María Xosé Rodríguez Álvarez (ed. lit.), 2020, ISBN 978-84-1319-267-3, págs. 434-437
• Idioma: inglés
• Enlaces
  • Texto Completo Libro
• Resumen

  • We investigate robust loss functions in statistical boosting, which is particularly suitable for high-dimensional data situations. To achieve robustness against outliers in the outcome variable we consider different robust losses. The stepwise boosting algorithm implicitly reweights the residuals in each iteration with the gradient of the loss function. For composite losses, e.g. the Huber and Bisquare loss, there is a cut-off value to choose. For this purpose, a fixed quantile for the amount of outliers is used that adapts this value in each iteration to the size of this residuals. As an application we investigate the performance of the boosting methods for various amounts of outliers in a high-dimensional ribo avin data set.