A Hybrid LSTM-DBN Approach for Automatic Epileptic Seizure Detection compared with Normal Human Activity using EEG Signals

More than 2% of people throughout the globe suffer from the neurodegenerative condition known as epilepsy. A sudden onset of convulsive seizures characterizes epilepsy, a condition of the neurological system. These seizures are brought on by an imbalance in the electrical processes of the brain. Neurostimulation devices possess the capability to pre-emptively intervene and prevent the onset of seizures. Its effectiveness can be improved by using techniques like advanced seizure forecast. The prompt detection of the preictal phase would facilitate timely activation of the neurostimulation device. In this work, a Deep Learning (DL) approach was utilized to detect the Epileptic Seizure (ES) before onset by utilizing Electroencephalogram (EEG) signals. Long Short-Term Memory (LSTM) model work focuses on evaluating the efficacy of several machine learning (ML) algorithms such as Logistic Regression (LR), Support Vector Machine (SVM) by applying them to Seizure recorded EEG data. Additionally, a Deep Belief Network (DBN) was employed to choose and interpret the captured dataset's most useful and efficient characteristics. The short name for this proposed methodology is LSTM-DBN. Ultimately, our objective is to identify the most efficient method for constructing a hybrid learning model. In this work, a public dataset was used to evaluate the ES such as CHB-MIT. These data are were collected from EEG signals. Comparing with conventional approaches of ML compared with DL techniques, LSTM-DBN had significantly got better performance with an accuracy of 99.07%