Patient-specific epileptic seizure prediction using correlation features

In this contribution, several classifiers are employed to study patient-specific epileptic seizure prediction quality using intracranial electroencephalogram signal (iEEG) for dogs and humans suffering from epilepsy. New approach to extraction of correlation-based features in sliding time window within the EEG epoch is proposed. Classification performance was evaluated by area under receiver operating characteristic curve (AUC). Influence of duration of time window on results of classification was studied. For epileptic seizure prediction in humans, best classification is showed by support vector machine classifier for time window T
_w
= 60 sec. (AUC=0.9349); for seizure prediction in dogs, highest obtained AUC is 0.9432 for SVM classifier and T
_w
= 30 sec.