Heart beat-to-beat intervals classification for epileptic seizure prediction

This work is devoted to the prediction of epileptic seizures using heart rate variability (HRV) characteristics. Several HRV features were extracted (statistical, spectral, histogram, polynomial approximation coefficients) for various durations of sliding time windows and various lengths of preictal intervals. The data from 14 subjects with generalized epileptic seizures was used. Support Vector Machine was exploited as a classifier. Leave-One-Group-Out validation, yielded the following values of classifier performance: AUC = 0.7622, sensitivity = 0.7252, specificity = 0.7252. These results indicate the possibility of seizure prediction using HRV characteristics. Our findings regarding positioning and sizing of used time windows, namely the imitations in finding optimal parameters between different subjects, could be used for further advancement of methods for epileptic seizure prediction using the heart rate variability characteristics.