EEG model forecasts fatigue-related reaction time 20 seconds ahead

Mental fatigue causes catastrophic accidents in sustained attention tasks like driving or air traffic control. Current neurophysiological systems detect ongoing performance decline but can't forecast lapses early enough for intervention. To address this, researchers Bo Sun and Liang Ma developed a model that predicts reaction time (RT) up to 20 seconds in advance using EEG functional connectivity. Thirty participants completed a sustained Psychomotor Vigilance Test (PVT) while 30-channel EEG was recorded. The team computed mutual information between electrode pairs as functional connectivity features, then trained a Random Forest regression model to forecast single-trial RTs across horizons from 0 to 20 seconds.

The model achieved immediate detection with a Root Mean Square Error (RMSE) of 23.75 ms and maintained high accuracy across forecasting horizons (RMSE = 24.07 ms). Interpretability analysis using SHAP and linear mixed effects models validated the approach and revealed distinct temporal biomarkers for fatigue. This study proves that behavioral performance can be reliably forecast 20 seconds before a lapse occurs, offering a practical method for proactive fatigue management in safety-critical environments—from long-haul trucking to surgical operations. The paper is available on arXiv under the subject Human-Computer Interaction.