Unified Robust Boosting
From MaRDI portal
Publication:83589
DOI10.48550/ARXIV.2101.07718arXiv2101.07718MaRDI QIDQ83589FDOQ83589
Publication date: 19 January 2021
Abstract: Boosting is a popular algorithm in supervised machine learning with wide applications in regression and classification problems. Boosting can combine a sequence of regression trees to obtain accurate prediction. In the presence of outliers, traditional boosting may show inferior results since the algorithm optimizes a convex loss function. Recent literature has proposed boosting algorithms to optimizing robust nonconvex loss functions. However, there is a lack of weighted estimation to indicate the outlier status of the observations. This article proposes an iteratively reweighted boosting algorithm combining robust loss optimization and weighted estimation, which is conveniently constructed with existing software. The output includes the weights as a valuable diagnostic to the outlier status of the observations. For practitioners interested in the boosting algorithm, the new algorithm can be interpreted as a method to tuning in observation weights, which can lead to a more accurate model. The R package irboost is demonstrated with publicly available data in various robust boosting approaches to generalized linear models, classification and survival data analysis.
Nonparametric robustness (62G35) Classification and discrimination; cluster analysis (statistical aspects) (62H30) Computational learning theory (68Q32)
Cited In (1)
This page was built for publication: Unified Robust Boosting
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q83589)
