Prediction of Absence Epileptic Seizures in Multichannel EEG Signals Using Hybrid Deep Learning Models: State of the Art, Challenges, and Future Perspectives

Absence epilepsy is a neurological disorder characterized by brief, sudden lapses in consciousness, whose subtle presentation often leads to underdiagnosis. Predicting these seizures is critical for preventing cognitive impairments and enabling timely intervention in children. This paper presents a critical state-of-the-art review focused on hybrid deep learning models-primarily combinations of convolutional and recurrent neural networks-applied to the prediction of seizures using multichannel EEG data, with a special focus on the challenges of absence epilepsy. This review analyzes trends in CNN-LSTM, CNN-GRU, and attention-based models. A key finding is that while promising, the field is hampered by significant challenges. This review critically highlights a major methodological pitfall: the over-reliance on patient-specific validation schemes, which inflates performance metrics and hinders the development of generalizable models for seizure prediction. Furthermore, the scarcity of well-labeled preictal datasets and the insufficient research focus on absence seizures are identified as primary barriers. The findings underscore the need for interpretable, robust, and clinically validated models for preictal state identification. Emerging approaches such as federated learning and explainable frameworks are discussed as key pathways to advance toward clinically deployable prediction systems.