Epileptic seizure detection from EEG signals using ensemble of type-2 neuro-fuzzy inference systems with...

Epileptic seizure detection from electroencephalography (EEG) signals is a challenging task due to the nonlinear, nonstationary, and uncertain nature of neural activity. This paper proposes a novel tri-domain epileptic seizure classification framework based on an ensemble of Interval Type-2 Neuro-Fuzzy Inference Systems (IT2FIS) with fuzzy rules optimized using the Grey Wolf Optimizer (GWO). The method integrates complementary EEG features extracted from the time, frequency, and time-frequency domains to provide a comprehensive representation of seizure-related dynamics. GWO is employed to initialize and optimize the antecedent and consequent parameters of the fuzzy rules, followed by a reconstruction-aided gradient-based fine-tuning process that enhances the discriminative capability of the IT2FIS architecture. An ensemble mechanism combines the outputs of three independently trained IT2FIS classifiers to improve robustness against signal variability. The framework is evaluated using 8 fuzzy rules per domain and a GWO population of 30 wolves, under both a stratified 5-fold cross-validation protocol for tuning and an unseen hold-out test set for final assessment. Experiments on the Bonn EEG dataset demonstrate strong performance, with the three-class classification (healthy, interictal, seizure) achieving 98.68% accuracy across individual feature domains and the multi-domain ensemble reaching near-perfect classification. Specifically, the multi-domain framework achieves near-perfect (98.68-100.00%) performance across three-class and all binary clinical splits, including 100% on the challenging E-CD scenario. Across all configurations, the framework achieves macro-F1 scores above 96%. The overall results show that the proposed hybrid fuzzy-optimization framework provides a highly accurate and robust solution for automated seizure detection with strong potential for clinical deployment.