A Self-Tuning ensemble approach for drift detection
Guilherme Yukio Sakurai, Bruno Bogaz Zarpel�ão, Sylvio Barbon · 2024
Processing data streams is challenging due to the need for mining algorithms to adapt to real-time drifts. Ensemble strategies for concept drift detection show promise, yet gaps in flexibility and detection remain. We propose the Self-tuning Drift Ensemble (StDE) method, which dynamically adapts ensemble structure to stream changes while maintaining a lightweight solution. StDE adjusts the number of base learners through a self-regulating voting system, achieving high detection accuracy. Experiments across various drift scenarios demonstrate the superior performance of our method compared to established baselines.