Online ensemble learning with abstaining classifiers for drifting and noisy data streams

Open Webpage

Bartosz Krawczyk, Alberto Cano

Published: 2018, Last Modified: 08 Mar 2025Appl. Soft Comput. 2018EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Lightweight and flexible extension of online ensembles with abstaining classifiers.•Dynamic selection of base classifiers to exploit their underlying diversity.•Improved recovery due to promoting classifiers correctly anticipating concept drifts.•Increased robustness to the presence of noise in data streams.•Thorough experimental study with analysis on 12 canonical and 120 noisy streams.