کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Interpretable K-Means-Based TSK Fuzzy Classification of EEG Signals for Seizure Detection

نویسندگان :

Arian Mesforoosh-M¹ Yeganeh Binafar² Mohammad-R Akbarzadeh-T³

1- دانشگاه فردوسی مشهد 2- دانشگاه فردوسی مشهد 3- دانشگاه فردوسی مشهد

کلمات کلیدی :

Seizure detection،Electroencephalography،Takagi-Sugeno-Kang fuzzy systems،K-means clustering،Interpretability

چکیده :

Epilepsy affects millions worldwide and poses a major neurological burden, particularly in resource-limited settings where access to long-term electroencephalography (EEG) monitoring is restricted. Although automated seizure detection has achieved remarkable accuracy using deep learning, most existing approaches suffer from limited interpretability and high computational complexity. This study proposes a clinically interpretable Type-1 Takagi–Sugeno–Kang fuzzy inference system in which K-means clustering is explicitly employed for data-driven fuzzy rule generation for EEG-based seizure detection. A compact set of 16 time- and frequency-domain features is hence extracted from preprocessed single-channel EEG signals, and predefined Gaussian membership functions with linear consequents learned via weighted least squares are employed. The resulting model provides explicit IF–THEN rules with low computational cost and full transparency. Experiments on the benchmark Bonn University EEG dataset across binary, ternary, and five-class classification tasks demonstrate accuracies up to 98.33% for binary problems, 96.00% for ternary classification, and 84.00% for the five-class scenario, with balanced macro-averaged F1-scores. Compared with deep learning approaches, the proposed framework achieves comparable performance while offering enhanced interpretability and potential suitability for real-time and wearable EEG applications.