EEG Signal Feature Extraction and Classification for Epilepsy Detection

Epilepsy is a neurological disorder of the central nervous system, characterized by sudden seizures caused abnormal electrical discharges in brain. Electroencephalogram (EEG) most common technique used for diagnosis. Generally, it done manual inspection EEG recordings active seizure periods (ictal). Several techniques have been proposed throughout years to automate this process. In study, we developed three different approaches extract features from filtered signals. The first approach was eight statistical directly time-domain signal. second approach, only frequency domain information applying Discrete Cosine Transform (DCT) signals then extracting two lower coefficients. last tool that combines both time and information, which Wavelet (DWT). Six wavelet families tested with their orders resulting 37 wavelets. decomposition levels were every wavelet. Instead feeding coefficients classifier, summarized them 16 features. extracted are fed classifiers k-Nearest Neighbor (k-NN), Support Vector Machine (SVM), Artificial Neural Network (ANN) perform binary classification scenarios: healthy versus epileptic(mainly interictal activity),and seizure-free ictal. We benchmark database, Bonn consists five sets. scenario, taken six combinations available data. While combinations. For detection (healthy vs epileptic), performed badly. Using DCT improved results, but best accuracies obtained DWT-based approach. detection, methods quite well. However, third method had performance better than many state-of-the-art terms accuracy. After carrying out experiments on whole signal, separated rhythms applied DWT Daubechies7(db7) feature extraction. observed close those recorded before can be achieved Delta rhythm scenario (Epilepsy detection) Beta (seizure detection).