Next-Gen Data Analytics and Intelligent Automation
Advanced Neural Architectures for EEG-Based Seizure Detection and Classification
Authors: Santushti Betgeri, Sangita M. Jaybhaye, Kaustubh Karne, Kartik Mehta, Ameya Kasetwar and Atharav Kasture
Publishing Date: 12-04-2026
ISBN: 978-81-975670-6-3
Abstract
Epileptic seizure classification and detection from electroencephalogram (EEG) data is an important field of investigation because seizure patterns pose considerable complexity and variable across patients. This paper investigates several different deep learning techniques such asTimesNet, LSTM, a hybrid model designed specifically for this purpose, transformer-based models such as FedFormer and Informer for seizure classification. Utilizing the TUH EEG Seizure Corpus (v1.5.4), we compare the effectiveness of these models for various seizure types, both intra- and inter-patient variability. The presented methods exhibit high classification performance, with some models outperforming state-of-the-art baselines. Our efforts also focus on model generalizability and real-time applicability, making them a step closer to clinically feasible seizure detection systems.
Keywords
EEG, Seizure Classification, Deep Learning, Transformer Models, 1D ResNet, Temporal Convolutional Network (TCN), TimesNet, Graph Neural Networks (GCN), LSTM,TUH Dataset, Epilepsy Detection, Neural Architecture Comparison.
Cite as
Santushti Betgeri, Sangita M. Jaybhaye, Kaustubh Karne, Kartik Mehta, Ameya Kasetwar and Atharav Kasture, "Advanced Neural Architectures for EEG-Based Seizure Detection and Classification", In: Kusum Kumari Bharti and Noor Firdoos Jahan (eds), Next-Gen Data Analytics and Intelligent Automation, SCRS, India, 2026, pp. 86-95. https://doi.org/10.56155/978-81-975670-6-3-8