DBTR-AGF: A Dual-Branch Transformer-Recurrent Network with Adaptive Gating for Epileptic Seizure Prediction

This study presents DBTR-AGF, a deep learning architecture combining temporal and spectral EEG analysis branches with adaptive fusion for predicting epileptic seizures. On the CHB-MIT dataset using leave-one-subject-out validation, the model achieved 99.02% accuracy with a 0.12/h false-positive rate and maintained robust performance even with reduced training data.