Early prediction of paroxysmal atrial fibrillation based on short-term heart rate variability

Atrial fibrillation (AF) is the most common arrhythmia type and its early stage is paroxysmal atrial fibrillation (PAF). PAF affects negatively the quality of life by causing dyspnea, chest pain, feeling of excessive fatigue, and dizziness. In this study, our aim is to predict the onset of paroxysmal atrial fibrillation (PAF) events so that patients can take precautions to prevent PAF events. We use an open data from Physionet, Atrial Fibrillation Prediction Database. We construct our approach based on the heart rate variability (HRV) analysis. Short-term HRV analysis requires 5-minute data so that each dataset was divided into 5-minute data segments. HRV features for each segment are calculated from time-domain measures and frequency-domain measures using power spectral density estimations of fast Fourier transform, Lomb-Scargle, and wavelet transform methods. Different combinations of these HRV features are selected by Genetic Algorithm and then applied to k-nearest neighbors classification algorithm. We compute the classifier performances by the 10-fold cross-validation method. The proposed approach results in 92% sensitivity, 88% specificity and 90% accuracy in the 2.5-7.5 min time interval priors to PAF event. The proposed method results in better classification performance than the similar studies in literature. Comparing the existing studies, we propose that our approach provide better tool to predict PAF events.