Epilepsy has become a commonplace neurological disorder that can be diagnosed in today’s world. However, early detection of epilepsy is very important to control the disease and it is still a challenging procedure in some cases. Investigation of early detection methods using advances in signal processing and machine learning techniques is getting a popular research area recently. Choiced features and used signal processing methods are very effective on the results of the detection procedures developed for the neurological disorders. This study is designed as a first step to decide which features and signal processing techniques are the most succesful for detection of neurological disorder like epilepsy. Time-domain and wavelet-domain features are analyzed in this work. Frequency-domain and time-frequency-domain features will be studied as the next step of this research. In this paper, we present a comparative investigation of machine learning methods for the detection of epilepsy using electroencephalogram (EEG) signals. We use statistical, auto-correlation based and wavelet transform based features of EEG signals to predict and detect epilepsy. In this work, states of epilepsy are classified using Naive Bayes (NB), J48 Decision Tree (DT), Random Forest (RF), Multilayer Perceptrons (MLP), k-Nearest Neighbor (k-NN), and Support Vector Machine (SVM) methods. Machine learning methods are applied to 500 EEG records in EEG database for epilepsy of the University of Bonn to classify them as normal - open eyes, normal - closed eyes, pre-ictal, inter-ictal, and ictal states. All the classification algorithms have been run in the Weka program. Results indicate that Random Forest, Multilayer Perceptrons, k-Nearest Neighbor (k-NN) with 10-fold cross-validation are the most successful algorithms with 92% accuracy.
