Synapse AI: A Multimodal Machine Learning Framework for Early Seizure Prediction Using Wearable Physiological Signals

Epilepsy is a neurological disorder characterized by recurrent seizures that often occur without warning, posing significant risks to patient safety and quality of life. While current treatments reduce seizure frequency for some individuals, reliable early seizure prediction remains a major challenge. This study developed Synapse AI, a prototype machine learning framework for early seizure prediction using electroencephalography (EEG) signals, intended to serve as the foundation for a future multimodal seizure prediction system integrating wearable physiological signals. EEG recordings from five cases in the publicly available CHB-MIT Scalp EEG Database were segmented into 10-second windows and labeled as preictal (5–30 minutes before seizure onset) or interictal. Statistical, time-domain, and frequency-domain features were extracted from each EEG window and used to train and compare three supervised machine learning models: XGBoost, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN). Model performance was evaluated using grouped train-test splitting, grouped cross-validation, and the metrics accuracy, precision, recall, F1 score, and ROC-AUC. Among the models evaluated, XGBoost achieved the strongest overall performance, obtaining an accuracy of 76.2%, recall of 84.6%, F1 score of 78.0%, and ROC-AUC of 85.2% on held-out test cases. These findings demonstrate the feasibility of using machine learning to identify preictal EEG patterns for early seizure prediction. Overall, this study establishes a proof-of-concept and provides a foundation for the development of more effective and accessible seizure prediction systems.