An Efficient EEG Channels-Selection Approaches For Epilepsy Seizure Prediction

Abstract

In this study, we are interested in the epilepsy seizures problem. Indeed, we used binary SVM to predict the ongoing seizures and multiclass SVM to predict different states of patients' epilepsy. Brain activity is used as an efficient source for predicting seizures, it's recorded in Electroencephalography (EEG) segments signal. We propose and compare in this paper, three ideas select channels: the highest frequency channels, the channels of the left part of the head, and the channels of the right part of the head. A features extraction stage is important to produce a rich and relevant dataset, in effect, 22 features are calculated for each segment of 5 min from EEG signal. A binary SVM is used to predict the ongoing seizures named pre-ictal, and a one-versus-all multi-class SVM is used to predict four classes (pre-ictal, ictal, inter-ictal, and post-ictal). A classification rate toward 97%, on the selected channels corpus, was achieved by SVM (binary and multiclass) with the majority of patients.