An efficient automated technique for epilepsy seizure detection using EEG signals

Epilepsy is a neurological disorder characterized by epileptic seizures. Epileptic seizure can be analyzed through the normal and abnormal activity of the brain. This abnormal activity can be observed only through the use of an efficient algorithm. The process of an efficient algorithm always uses signal processing in which an epileptic signal can be considered as an input signal. This paper introduces a technique to detect epileptic signal and to compare the characteristics of the brain signals at different stages. Our algorithm is based on signal processing techniques to detect epilepsy in the EEG signal. The signal processing starts with sampling the signal at 178.6 Hz so that the signal operating frequency follows oversampling criteria. The sampled signal is given to the designed filter so that the unwanted noise can be removed and the signal is ready to be decomposed. Then, the signal is decomposed at five different signal levels so that its frequency spectrum is reduced to less than 200 Hz using different wavelet filters at each level. In the feature extraction, we have used signal features rather than statistical features so that we can still rely on time domain and frequency domain features for an EEG signal. These features are classified using Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Artificial Neural Networks (ANN) to detect the epilepsy in the EEG signal. The results were demonstrated for different sets of brain signal that show the normal behavior of the brain signals and epileptic behavior of the signal with seizure. A comparison of our work with the present traditional methodologies proves that our algorithm is more efficient in detecting epilepsy.