Atrial fibrillation (AF) is the most clinically diagnosed arrhythmia, as its prevalence increases with age, and its initial stage is paroxysmal atrial fibrillation (PAF). This pathology usually triggers hemodynamic disorders that can generate cerebrovascular accidents (CVA), causing morbidity and even death. The aim of this study is to predict the occurrence of PAF episodes in order to take precautions to prevent PAF episodes. The PhysioNet AFPDB prediction database was used to extract 77 heart rate variability (HRV) features using time domain, geometrical analysis, Poincaré plot, nonlinear analysis, detrended fluctuation analysis, autoregressive modeling, fast Fourier transform (FFT), Lomb-Scargle periodogram, wavelet packet transform (WPT) and bispectrum measurements. The number of features was reduced using the near-zero value, correlation, and recursive feature elimination (RFE) methods for time windows of 1, 2, 5, 10, and 30 min. Feature selection was performed using backwards selection, genetic algorithm, analysis of variance (ANOVA), and non-dominated sorting genetic algorithm (NSGA-III) methods, and then random forest, conditional random forest, k-nearest neighbor (KNN), and support vector machine (SVM) classification algorithms were applied and evaluated using 10-fold cross-validation. The proposed method achieved a precision of 93.24% with a 5-minute window and 89.21% with a 2-minute window, improving performance in predicting PAF when compared with similar studies in the literature.

Atrial fibrillation (AF) is the most clinically diagnosed arrhythmia in cardiac pathology. The incidence of AF begins at a very early age and its initial state is paroxysmal atrial fibrillation (PAF). This type of heart disease can be detected and predicted by analyzing the spectrogram of a surface electrocardiogram (ECG) signal. In many studies, different ECG signal formats and convolutional neural network (CNN) architectures have been used. However, the lack of good signal preprocessing or signal adequacy may have affected the accuracy, especially on short-term ECG signals. In this study, we analyzed a preprocessed ECG signal, determined the optimal set to predict PAF, and evaluated the accuracy using ECG signals of different durations. The PAF Prediction Challenge–PhysioNet database was used to extract spectrograms in 30-sec and 5-sec windows for two classes (Normal, PAF) and 3 classes (Normal, Close-AF, Distant-AF). Then, the AlexNet architecture was used. The proposed method achieved a two-class accuracy of 99.92% with a 30-sec window and 99.42% with a 5-sec window, improving the PAF prediction performance compared with similar works. In addition, the three-class accuracies were 96.92% and 97.43% with windows of 30-sec, and 5-sec, respectively. These results prove the efficacy of the method for the early diagnosis of PAF, even based on short-term ECG signals.