DynaQ: online learning from imbalanced multi-class streams through dynamic sampling

Abstract Online supervised learning from fast-evolving data streams, particularly in domains such as health, the environment, and manufacturing, is a crucial research area. However, these often experience class imbalance, which can skew distributions. It essential for online algorithms to analyze large datasets real-time while accurately modeling rare or infrequent classes that may appear bursts. While methods have been proposed handle binary there lack of attention multi-class imbalanced settings with varying degrees imbalance evolving streams. In this paper, we present Dynamic Queues (DynaQ) algorithm fill knowledge gap. Our approach utilizes batch-based resampling method creates an instance queue each balance number instances. We maintain threshold remove older samples during training. Additionally, dynamically oversample minority based on one four rate parameters: recall, F1-score, $$\kappa _m$$ κ m , Euclidean distance. consists ensemble uses sliding windows soft voting schema incorporating drift detection mechanism. experimental results demonstrate superiority DynaQ over state-of-the-art methods.