EEG Data Classification by Hybrid Deep Learning for Epileptic Seizure Prediction

Epileptic seizures pose a considerable health challenge, affecting a large population globally. Conventional methods for detecting seizures typically involve neurologists' manual interpretation of electroencephalogram (EEG) signals, which is labor-intensive and prone to inaccuracies. Most existing research on epilepsy prediction using EEG signals focuses on patient-specific approaches, where prediction models are individually tailored and trained for each patient. While effective, this method requires extensive personalized data and may not generalize well across different patients. In contrast, this work aims to develop a multi-subject approach to address the problem of generalizing to new patients, where a single, robust prediction model can be applied across diverse patient populations. This strategy simplifies the prediction process and holds promise for broader clinical applicability, providing a more scalable solution for epileptic seizure prediction. The proposed approach uses hybrid deep learning by merging ID-CNN blocks for spatial features, Bi-LSTM for temporal features, and an attention mechanism. Our results highlight the potential of hybrid deep learning to surpass the limitations of traditional approaches and provide a robust framework for epileptic seizure prediction.