Classification of epileptic EEG signals based on machine learning methods

According to World Health Organization (WHO) analyses, epilepsy is the most common neurological disorder in the world. This disorder, that can affect people of all ages, leads to loss of consciousness, movement disorder, and like other cognitive functions causes temporary involuntary situations. In order to make a diagnosis of Epilepsy, which is also called as sara, the electrical signal in the brain is recorded by non-invasive or intracranial EEG methods. Due to the wide variety of morphologies of epileptic EEG signals, the evaluation them by neurophysiologists could be both time-consuming and misleading. Therefore, studies have been carried out to develop integrated models of diagnosis of epilepsy for local health systems. There are machine learning-based approaches in the literature to assist specialists in diagnosing epilepsy from EEG signals. In this thesis, the aim is to develop a method which will be beneficial to medical diagnostic systems and reduce the transaction cost in seizure detections. Three separate studies were performed using the open-access EEG epileptic dataset that have carried out by the University of Bonn University Epileptology Department. In the studies, features that obtained by different methods have classified based on Extreme Learning Machines, Support Vector Machines, K-Nearest Neighbour, and Decision Trees. In addition, for the system which the highest performance results was obtained by, ROC curve (Receiver Operation Characteristics), that commonly used in classification assessment in diagnostic test, was used to analyse classifier's performance evaluation. Finally, obtained results compared with the studies in the literature that have been used.