A Multi-Dimensional CNN-Bi-GRU for IoT-Based Brain–Computer Interface in Early Epileptic Seizure Detection

Abstract The study focuses on seizure detection using EEG data from Mendeley. An early-alert IoT-BCI system is designed to simulate real-time support for patients during seizures. The proposed Multi-Dimensional CNN-Bi-GRU (MDCBG) outperforms hybrid deep learning models, achieving 97.43% accuracy, surpassing baseline EEGNet (92.17%) and CTNET (85.11%), along with models evaluated through ablation studies on seizure vs. non-seizure prediction. The proposed model, along with other models like Bi-GRU with attention, Bi-LSTM-GRU, and XGBoost, also performs well on classifying various types of seizures. SHAP analysis shows Channel 5 contributes most to predictions. An IoT-based automation system is simulated on seizure detection for triggering micro devices near the patient’s environment. This approach supports early seizure warning and guides home-automation strategies to assist patients.