The Investigation of Heart Rate Variability for Paroxysmal Atrial Fibrillation Detection

Paroxysmal atrial fibrillation ( PAF) i s t he initial phase of atrial fibrillation ( AF), often progressing stealthily to the chronic stage due to the absence of noticeable symptoms. Hence, timely identification of PAF is pretty necessary. This study proposes an automated machine learning-based PAF detection algorithm utilizing a single-lead electrocardiogram signal. A total of 25 features are extracted from 1-minute segments and the optimal feature set, selected by deploying the minimum redundancy maximum relevance algorithm, is used to train decision tree (DT) and random forest (RF) classifiers. The training and testing stages included 43 subjects, and subjectwise 10-fold cross-validation was performed. RF outperforms the DT classifier a chieving 91.94% accuracy, 91.75% sensitivity, and 91.47% F1 score. The higher accuracy using shorter ECG segments remarks the significance of the proposed model for AF monitoring.