Feature selection and classification of Epilepsy from EEG signal

Abstract

Epilepsy is one of the most common disorders in a neurological system that affects almost 50 million population of all ages worldwide. Epilepsy denotes “seizure disorders” which is characterized by unpredictable chronic seizures. Epilepsy is a spectrum condition with a broad range of seizure types varying from person-to-person & it is commonly diagnosed by EEG, Magnetic Resonance Imaging, and fMRI and also by using Magnetoencephalography. The traditional method of analyzing EEG is based on using strip charts to visually analyze the EEG activity which is a laborious and time-consuming task. Therefore in this work, an automated diagnosis method using ANN was designed to classify Epilepsy from EEG based on different stages of EEG signal levels (Ictal, Inter-ictal, Pre-ictal). After preprocessing the signal, features like mean, variance, skewness, kurtosis, standard deviation are extracted. Then for accuracy in classification & to reduce the dimensionality of dataset feature ranking was implemented. Finally, neural networks was implemented to classify Epilepsy based on risk levels & this method of classification provides an accuracy of 96.9%.