EEG-Based Epileptic Seizure Prediction Using Variants of the Long Short Term Memory Algorithm

Abstract

One of the most widespread neurological disorders worldwide is epilepsy. Seizures that are caused by sudden aberrant electrical activity in the patient’s brain, causing unpredictable episodes, are called epileptic seizures. Epileptic seizure patient’s life can be significantly impacted by early detection of seizure. For the prediction of seizures based on zero-crossing intervals of examination of the scalp, an approach known as electroencephalograms (EEGs) is introduced. EEG signals are examined to anticipate seizures and prevent unwarranted life risks. Deep Learning (DL) has been used in this research as it can automatically generate hierarchical representation from unprocessed EEG data, making it possible to quickly uncover complicated patterns and information associated with brain activity. This required preprocessing EEG scalp recordings, automatic feature extraction, and classification. In this paper, we proposed the Long Short Term Memory (LSTM) variants, Bi-LTSM, vanilla LSTM, and stacked LSTM, and compared the results with GRU, MLP, and DCNN for epileptic seizure prediction. We compared these models using the CHB-MIT dataset to improve accuracy, sensitivity, and specificity for predicting epileptic seizures. The results show that the Bi-LSTM algorithm performed better than the other proposed algorithms in terms of evaluation metrics.