Epileptic Seizure Detection using Deep Ensemble Network with Empirical Wavelet Transform

Abstract

Epileptic seizure attack is caused by abnormal brain activity of human subjects. Certain cases will lead to death. The detection and diagnosis is therefore an important task. It can be performed either by direct patient activity during seizure or by electroencephalogram (EEG) signal analysis by neurologists. EEG signal processing and detection of seizures using machine learning techniques make this task easier than manual detection. To overcome this problem related to a neurological disorder, we have proposed the ensemble learning technique for improved detection of epilepsy seizures from EEG signals. In the first stage, EEG signal decomposition is done by utilizing empirical wavelet transform (EWT) for smooth analysis in terms of sub-bands. Further, features are extracted from each sub. Time and frequency domain features are the two categories used to extract the statistical features. These features are used in a stacked ensemble of deep neural network (DNN) model along with multilayer Perceptron (MLP) for the detection and classification of ictal, inter-ictal, and pre-ictal (normal) signals. The proposed method is verified using two publicly available datasets provided by the University of Bonn (UoB dataset) and Neurology and Sleep Center - New Delhi (NSC-ND dataset). The proposed algorithm resulted in 98.93 % and 98 % accuracy for the UoB and NSC-ND datasets, respectively.