Epileptic Seizure Detection via EEG Signals with Feature Extraction Technique Based on Cubic Spline Interpolation (In English)

Epilepsy is a common neurological disease nowadays. Epilepsy causes repeated seizures with various time intervals in patients. During seziure, patients generally lose their consciousness, hence, the risks of self-harm and self-injury increases. One of the most common techniques used for epileptic seizure diagnosis is Electroencephalogram (EEG). In this study, a method based on feature extraction with cubic spline interpolation is proposed for epileptic seizure detection. EEG dataset published by Bonn University is employed. In the first stage, EEG signals are preprocessed and upper and lower envelopes are extracted by cubic spline interpolation. In the second stage, to obtain features that represent the common attributes of the upper and lower envelopes, the Manhattan distance between envelopes is computed. In the final stage, for the classification, features obtained by Manhattan distance are supplied to Support Vector Machines (SVM), K-Nearest Neighbors (K-NN) and Decision Trees (DT) classifiers. The classification is made between classes that contain healthy samples or seizure-free epilepsy samples and class that contains samples of epilepsy with seizure. The highest accuracy rate is obtained with SVM for classes A-E, which is 99%.