MOA Concept Drift Active Learning Strategies for Streaming Data

Indre Zliobaite, Albert Bifet, Geoff Holmes, Bernhard Pfahringer
Proceedings of the Second Workshop on Applications of Pattern Analysis, PMLR 17:48-55, 2011.

Abstract

We present a framework for active learning on evolving data streams, as an extension to the MOA system. In learning to classify streaming data, obtaining the true labels may require major effort and may incur excessive cost. Active learning focuses on learning an accurate model with as few labels as possible. Streaming data poses additional challenges for active learning, since the data distribution may change over time (concept drift) and classifiers need to adapt. Conventional active learning strategies concentrate on querying the most uncertain instances, which are typically concentrated around the decision boundary. If changes do not occur close to the boundary, they will be missed and classifiers will fail to adapt. We propose a software system that implements active learning strategies, extending the MOA framework. This software is released under the GNU GPL license.

Cite this Paper

BibTeX
@InProceedings{pmlr-v17-zliobaite11a, title = {MOA Concept Drift Active Learning Strategies for Streaming Data}, author = {Zliobaite, Indre and Bifet, Albert and Holmes, Geoff and Pfahringer, Bernhard}, booktitle = {Proceedings of the Second Workshop on Applications of Pattern Analysis}, pages = {48--55}, year = {2011}, editor = {Diethe, Tom and Balcazar, Jose and Shawe-Taylor, John and Tirnauca, Cristina}, volume = {17}, series = {Proceedings of Machine Learning Research}, address = {CIEM, Castro Urdiales, Spain}, month = {19--21 Oct}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v17/zliobaite11a/zliobaite11a.pdf}, url = {https://proceedings.mlr.press/v17/zliobaite11a.html}, abstract = {We present a framework for active learning on evolving data streams, as an extension to the MOA system. In learning to classify streaming data, obtaining the true labels may require major effort and may incur excessive cost. Active learning focuses on learning an accurate model with as few labels as possible. Streaming data poses additional challenges for active learning, since the data distribution may change over time (concept drift) and classifiers need to adapt. Conventional active learning strategies concentrate on querying the most uncertain instances, which are typically concentrated around the decision boundary. If changes do not occur close to the boundary, they will be missed and classifiers will fail to adapt. We propose a software system that implements active learning strategies, extending the MOA framework. This software is released under the GNU GPL license.} }
Endnote
%0 Conference Paper %T MOA Concept Drift Active Learning Strategies for Streaming Data %A Indre Zliobaite %A Albert Bifet %A Geoff Holmes %A Bernhard Pfahringer %B Proceedings of the Second Workshop on Applications of Pattern Analysis %C Proceedings of Machine Learning Research %D 2011 %E Tom Diethe %E Jose Balcazar %E John Shawe-Taylor %E Cristina Tirnauca %F pmlr-v17-zliobaite11a %I PMLR %P 48--55 %U https://proceedings.mlr.press/v17/zliobaite11a.html %V 17 %X We present a framework for active learning on evolving data streams, as an extension to the MOA system. In learning to classify streaming data, obtaining the true labels may require major effort and may incur excessive cost. Active learning focuses on learning an accurate model with as few labels as possible. Streaming data poses additional challenges for active learning, since the data distribution may change over time (concept drift) and classifiers need to adapt. Conventional active learning strategies concentrate on querying the most uncertain instances, which are typically concentrated around the decision boundary. If changes do not occur close to the boundary, they will be missed and classifiers will fail to adapt. We propose a software system that implements active learning strategies, extending the MOA framework. This software is released under the GNU GPL license.
RIS
TY - CPAPER TI - MOA Concept Drift Active Learning Strategies for Streaming Data AU - Indre Zliobaite AU - Albert Bifet AU - Geoff Holmes AU - Bernhard Pfahringer BT - Proceedings of the Second Workshop on Applications of Pattern Analysis DA - 2011/10/21 ED - Tom Diethe ED - Jose Balcazar ED - John Shawe-Taylor ED - Cristina Tirnauca ID - pmlr-v17-zliobaite11a PB - PMLR DP - Proceedings of Machine Learning Research VL - 17 SP - 48 EP - 55 L1 - http://proceedings.mlr.press/v17/zliobaite11a/zliobaite11a.pdf UR - https://proceedings.mlr.press/v17/zliobaite11a.html AB - We present a framework for active learning on evolving data streams, as an extension to the MOA system. In learning to classify streaming data, obtaining the true labels may require major effort and may incur excessive cost. Active learning focuses on learning an accurate model with as few labels as possible. Streaming data poses additional challenges for active learning, since the data distribution may change over time (concept drift) and classifiers need to adapt. Conventional active learning strategies concentrate on querying the most uncertain instances, which are typically concentrated around the decision boundary. If changes do not occur close to the boundary, they will be missed and classifiers will fail to adapt. We propose a software system that implements active learning strategies, extending the MOA framework. This software is released under the GNU GPL license. ER -
APA
Zliobaite, I., Bifet, A., Holmes, G. & Pfahringer, B.. (2011). MOA Concept Drift Active Learning Strategies for Streaming Data. Proceedings of the Second Workshop on Applications of Pattern Analysis, in Proceedings of Machine Learning Research 17:48-55 Available from https://proceedings.mlr.press/v17/zliobaite11a.html.

Related Material