Seizure detection based on autoregressive modeling

This paper considers the use of autotoregressive (AR) modeling of Electroencephalogram (EEG) signals to discriminate between normal and epileptic EEG signals on one hand and to descriminate between seizure and seizure-free EEG signals on the other hand. Each epoch of EEG signal is modeled by an AR model of order P. Then, the obtained P AR coefficients are used in training and testing of a support vector machine (SVM) classifier. The optimal AR model order is investigated. The method is tested against a widely used EEG database and results show a classification accuracy of 100% when considering normal and epileptic EEG signals and a classification accuracy of 96.54% when considering seizure and seizure-free EEG signals. The obtained results are along with those obtained by state of the art EEG signal classifiers.