This study investigates the detection of Atrial Flutter in highly imbalanced ECG datasets. Using the XGBoost classifier, multiple resampling techniques, anomaly detection methods and Active Learning strategies were evaluated. Synthetic data generated by GANs significantly improved minority-class recognition, enhancing recall and F1-score. With emphasis on Active Learning algorithms, certainty-based and FIMPS, iteratively refined the model on the most uncertain samples. Results demonstrate that combining GAN augmentation with Active Learning outperforms traditional resampling methods. This approach highlights the potential for more accurate and efficient automated ECG analysis.

This work was (partially) supported by ISTAR-Iscte Pluriannual Projects UIDB/04466/2025 and UIDP/04466/2025

imbalanced learning,Atrial flutter,Active learning,ECG,GAN