Classification of Epileptic Seizure in EEG Signal Using ANFIS and Optimized Statistical Features by Genetic Algorithm

Epilepsy attributed to a known brain disorder in which nerve cells receive excessive abnormal neuronal firings. Identification and classification of the epileptic and non-epileptic electroencephalographic signals play an important role in clinical investigations. In this study, we used in vivo EEG signals recorded from epilepsy patients during the seizure and seizure-free intervals. We proposed an EEG analysis of seizure detection, based on a cascade of four stages: 1) stationary parts of EEG signals were decomposed into five time-frequency sub-bands using discrete wavelet transform (DWT) .2) statistical features were extracted from wavelet coefficients in each frequency band. 3) The feature sets were passed through a genetic algorithm to extract optimized number of features along each time pointed. 4) The overall accuracy of the classification and performance of the ANFIS network using selected features were presented. Our results reached 99.4% accuracy by just “mean” statistical feature. This paper verified the performance and usefulness of such a cascade system for diagnosing seizure events once there are suspected clinical symptoms of epileptic specifically in newborns.