Fatigue is a feeling of tiredness or weakness that can be physical, mental or both. Interpretable machine learning methods can be used for inference while having the dimension of time series prediction for developing intelligent health systems to manage fatigue. The study explored forecasting perception and performance scores associated with fatigue manifestation using cardiac activity with generalized and personalized models at a forecast interval of 10-minutes during a cognitively fatiguing task. Participants underwent a 2-hour cognitively fatiguing working memory task with subjective fatigue responses obtained every 10-minutes. Participants performance was calculated over the 10-minute interval. Performance labeled models had the lowest mean absolute error for forecasting in both generalized and personalized models using Gradient Boosting Regression. This provides the ability to forecast performance decrease due to fatigue and to generate fatigue mitigation interventions to reduce fatigue-related injuries.
