Predicting Epileptic Seizures using Ensemble Method

Abstract

Medication or surgical treatment is the techniques used for people diagnosed with epilepsy, but these procedures are not completely effective. Nevertheless, therapeutic method can be employed in the prediction of the seizure at an early stage. This is because it has been made known through research that the irregular activity in the brain begins a few minutes before the seizure start, the condition normally referred to as preictal state, which is known as a preictal state. Different Deep learning algorithms have been applied to detect seizures in Electroencephalogram (EEG) data. Though, several of the Epileptic Seizures (ES) prediction models have suffered from a lack of reliability and reproducibility due to the flaw in setting up a model to classify seizure prediction. The use of deep learning techniques is proposed to set up an ensemble model that will predict epileptic seizures. In the proposed method, Scalp EEG signals are used and they were gotten from the following repositories, TUG EEG Corpus, CHB-MIT, and GitHub EEG Repository later preprocessed. Univariate features were extracted from the preprocessed signal using signal mapping. The three deep learning techniques, Sparse Autoencoder (SAE), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) are independently trained with the data obtained from the feature extraction process. Majority Voting and Fusion Function are used to develop the ensemble model. 200 subjects of scalp EEG dataset were fed into the proposed system to test for scalability, the results successfully show an achievement of an average accuracy, sensitivity, and specificity of 97.4%, 96.1%, and 98% respectively.