Epileptic Seizure Prediction Using 1D-MobileNet

Automatic seizure prediction is helpful to Epilepsy patients to identify the possibility of seizure events momentarily. This could mitigate the risk faced by them while driving, climbing a staircase, or in similar situations in public. Since the olden days, clinicians have widely used electroencephalography (EEG) for examining epilepsy patients' brain activities. Yet, manually inspecting seizure episodes from EEG recordings is a time-consuming operation that may be reduced with the assistance of an automated seizure prediction system. The proposed work focuses on implementing a seizure detection mechanism, which inturn benefit the lives of both Epilepsy sufferers and clinicians. To do so, seizure phases must be identified, and a 1D-MobileNet architecture is used in this study to classify the ictal and preictal stages of the seizure. Prior to feeding the EEG data into the model, we employed fewer channels and data augmentation based on the sliding window technique. We assessed the proposed model on the CHB-MIT dataset, which yielded an accuracy of 98.50%, average precision of 99.00%, recall of 98.06%, and F1-score of 98.00%.