Detection of epileptic seizure using Support Vector Machine Classifier - extracted features from EEG signals

Abstract

Epilepsy is the most common neurological issue in people after stroke. Around 40 or 50 million individuals on the planet endure epilepsy. Epilepsy is characterized by an irregular seizure in which abnormal electrical activity in the mind causes adjusted recognition or conduct. The most commonly used test for detecting Epilepsy is EEG - which stands for Electroencephalogram. In this thesis, we tried to develop an automated system using machine learning that can detect epileptic seizure. We cropped one hour of pre-seizure and post-seizure signal and extracted features from it. We used Fast Fourier Transformation to make our data easier to process and applied Power Spectrum Density (PSD) to calculate energy from it. Finally we used Support Vector Machine (SVM) to classify among these data to differentiate between seizure and non-seizure. We have managed to achieve 89% accuracy using this method on the 23 cases that we had in our dataset.