Learning under Concept Drift: A Review

Learning under Concept Drift: A Review

13 Apr 2020 | Jie Lu, Fellow, IEEE, Anjin Liu, Member, IEEE, Fan Dong, Feng Gu, João Gama, and Guangquan Zhang
This paper reviews over 130 high-quality publications on concept drift, analyzing current methodologies and techniques, and establishing a framework for learning under concept drift with three main components: drift detection, understanding, and adaptation. It lists and discusses 10 synthetic datasets and 14 benchmark datasets for evaluating learning algorithms. The paper also covers research directions in concept drift. The framework includes drift detection (identifying drift occurrence), understanding (determining when, how, and where drift occurs), and adaptation (reacting to drift). The paper discusses various drift detection methods, including error rate-based, data distribution-based, and multiple hypothesis tests. It also covers drift understanding, focusing on the time, severity, and regions of drift. Drift adaptation strategies include simple retraining, ensemble retraining, and model adjustment. The paper highlights the importance of concept drift in data streams and provides a comprehensive review of current research and trends in the field.This paper reviews over 130 high-quality publications on concept drift, analyzing current methodologies and techniques, and establishing a framework for learning under concept drift with three main components: drift detection, understanding, and adaptation. It lists and discusses 10 synthetic datasets and 14 benchmark datasets for evaluating learning algorithms. The paper also covers research directions in concept drift. The framework includes drift detection (identifying drift occurrence), understanding (determining when, how, and where drift occurs), and adaptation (reacting to drift). The paper discusses various drift detection methods, including error rate-based, data distribution-based, and multiple hypothesis tests. It also covers drift understanding, focusing on the time, severity, and regions of drift. Drift adaptation strategies include simple retraining, ensemble retraining, and model adjustment. The paper highlights the importance of concept drift in data streams and provides a comprehensive review of current research and trends in the field.
Reach us at info@study.space
Understanding Learning under Concept Drift%3A A Review