Detection of Persistent Atrial Fibrillation Using ECG Signal

Persistent atrial fibrillation (PersAF) is a category of atrial fibrillation (AF) that endures for approximately a week and can easily revert back to a normal rhythm. Nonetheless, if left untreated, it can progress to chronic AF due to increasing complexity. Thus, the timely identification of PersAF necessitates more effective automatic detection algorithms. This research introduces a machine learning-driven automated algorithm designed to detect PersAF using a single-lead electrocardiogram (ECG) signal. By analyzing various time, frequency, and entropy features extracted from 10-second ECG segments, the best combination of features selected by deploying feature selection algorithms was used to train the k-nearest neighbor (KNN), decision tree (DT), and random forest (RF) classifiers. The training and testing phases involved 105 subjects, and the model's performance was validated using the 10-fold cross-validation technique. Among the classifiers, RF demonstrates the highest efficacy, achieving 95.40 ± 2.28% accuracy, 96 ± 2.81% sensitivity, 93.42 ± 5.81% specificity, and 0.94 ± 0.04 F1 score. The proposed method is thus shown to predict PersAF incidents with notable precision using shorter ECG segments.