A Comparison of Classifiers for Epileptic Seizure Prediction Based on Heart Rate Variability

Epilepsy is a neurological disorder characterized by recurrent seizures due to abnormal discharges in cortical networks of the brain. A seizure prediction method with a low false-positive rate in a high confidence interval and without side effects may improve patients’ quality of life. Heart rate variability (HRV) analysis is among the most promising approaches for seizure prediction. This method indirectly assesses the behavior of the autonomic nervous system (ANS) through cardiac rhythm activity. Artificial intelligence (AI) classifiers may predict seizures and distinguish the different phases in ECG signals. This work evaluated several classifiers for seizure prediction and studied them in terms of computational cost for training, sensitivity, accuracy, false-positive rate (FPR), and their suitability for wearable applications using the HRV approach. Relied on the results, the Support Vector Classifier (SVC) obtained the best set of scores, including the highest accuracy, 97.57%, as well as the second-highest Sen, Spe, and NPV scores, 97.70%, 97.51%, and 98.83%, respectively for preictal periods, considering an evaluation of 14.08 h from six patients’ ECG data.