Gml-PAF: A Generalizable Machine Learning Algorithm for Paroxysmal Atrial Fibrillation Detection based on Short-Term Inter-Beat Intervals

Reliable detection of paroxysmal atrial fibrillation (PAF) poses a significant challenge. We propose a generalizable machine learning (ML) algorithm for PAF detection (Gml-PAF) that uses 21-beat inter-beat intervals (IBI). Gml-PAF employs a model-agnostic framework integrating model selection, feature selection, and hyperparameter tuning. It is trained and evaluated across 16 PhysioNet electrocardiogram (ECG) databases, demonstrating robust cross-database generalization. In independent tests, it achieves F1 scores of 0.747–0.987 and AUC values of 0.933–0.999. The algorithm matches deep learning (DL) performance with longer IBI sequences and surpasses conventional ML methods, confirming its strong utility for wearable screening.