A Decision Support System for Prediction of Paroxysmal Atrial Fibrillation based on Heart Rate Variability Metrics

Abstract

Paroxysmal atrial fibrillation (PAF) is a temporary arrhythmic condition which is often a precursor of permanent/chronic atrial fibrillation. As frequent PAF events may easily lead to serious heart conditions, such as stroke, arterial embolism, it is propitious to have an early warning system. To this end, we propose an automated system for early prediction of PAF events based on statistical and nonlinear features extracted from heart rate variability (HRV) signal. We compute multiscale symbolic entropy, visibility graph-based complexity measures and three time-domain measures from the HRV signal. Out of them, the independent discriminative features are selected by Wilcoxon signed-rank test and correlation assessment. Finally, the selected features are applied to support vector machine (SVM), naive Bayes, and logistic regression classifiers to obtain the best prediction model. We achieve the best prediction results using radial basis function based SVM classifier with sensitivity, specificity and accuracy of 96.15%, 97.06%, 96.80% respectively, from the segments 5-10 mins before the onset of PAF events.