Classification of Epileptic EEG Data by Using Ensemble Empirical Mode Decomposition

In this study, our aim is to distinguish pre-seizure and seizure data from epileptic EEG signals using Ensemble Empirical Mode Decomposition (EEMD) and various classifiers. For this purpose, epileptic EEG data from 13 epileptic patients have been recorded using surface electrodes at Izmir Katip Celebi University School of Medicine, Neurology Department. First, EEG signals are divided into two parts as pre- seizure and seizure and decomposed into intrinsic mode functions (IMFs) using EEMD. Total power and higher order frequency moments are calculated as features from the first IMF by periodogram method. Extracted features are classified using Naive Bayes, K-nearest neighbors and Linear Discriminant Analysis methods. From the obtained classification results, it is seen that the Naive Bayes classifier outperforms the K-nearest neighbor and Linear Discriminant Analysis classification methods in pre-seizure and seizure data discrimination with maximum 100% and minimum 77% accuracy.