Keywords
Migraine, EEG, wearables, ML, classification

Technologies
MNE, scikit-learn, Empatica, ANT Neuro, BIDS

Migraine is a recurrent neurological disorder affecting approximately 12% of adults worldwide and ranking as the second leading cause of disability globally. Current treatments are reactive, with triptans losing efficacy within hours of headache onset. This project investigates whether migraine attacks can be predicted from physiological signals recorded before symptom onset, using data from the MigMark study conducted at Tyndall. Laboratory EEG and continuous Empatica E4 wearable data were collected from 16 participants across 72 recording sessions.

Sessions were labelled as pre-ictal or inter-ictal based on self-reported migraine diaries. Spectral band power, relative power and hemispheric asymmetry features were extracted from EEG recordings, while heart rate, electrodermal activity and skin temperature features were derived from wearable data. Three parallel tracks were evaluated: EEG, Empatica and multimodal. All experiments used participant level data splitting with LOSO-CV and a test set to ensure honest generalization estimates.