Unified Robust Boosting
share
Boosting is a popular machine learning algorithm in regression and classification problems. Boosting can combine a sequence of regression trees to obtain accurate prediction. In the presence of outliers, traditional boosting, based on optimizing convex loss functions, may show inferior results. In this article, a unified robust boosting is proposed for more resistant estimation. The method utilizes a recently developed concave-convex family for robust estimation, composite optimization by conjugation operator, and functional decent boosting. As a result, an iteratively reweighted boosting algorithm can be conveniently constructed with existing software. Applications in robust regression, classification and Poisson regression are demonstrated in the R package ccboost.
READ FULL TEXT