Fine-Tuned and Optimized Transformer-Based Model for EEG Seizure Detection 

Abstract

Epilepsy, a widely recognized neurological disorder, results from irregularities in the transmission of electrical impulses among neurons in the brain. Over the last two decades, significant efforts have been made by researchers and clinicians to develop effective methods for its early detection and management. The electroencephalogram (EEG), a non-invasive tool used to monitor brainwave activity, has become a central device in seizure diagnosis. With recent advances, EEG-based analysis is increasingly supported by machine learning and metaheuristic optimization approaches to enhance diagnostic accuracy and efficiency. This research proposes an optimized framework for seizure detection that leverages a Regularized Extreme Learning Adaptive Neuro-Fuzzy Inference System (R-ELANFIS) as the primary classifier. To reduce computational overhead and improve solution accuracy, a hybrid metaheuristic algorithm combining Particle Swarm Optimization (PSO) and Parrot Optimization (PO) is applied to fine-tune the model. The Bonn University EEG dataset, known for its reliable short-term seizure recordings, is used to evaluate system performance. Key classification metrics such as accuracy, sensitivity, and specificity reflect the model’s strong predictive capability with accuracy reaching up to 98.3%. The proposed method demonstrates the potential for high-performance EEG-based seizure detection paving the way for future integration with edge computing devices to support remote clinical diagnostics and continuous monitoring in real-world healthcare applications.