Purpose: Predicting in-game heart rate (HR) in dynamic team sports remains challenging due to limitations of models developed under controlled, laboratory-based conditions. We address this by developing a context-aware framework that forecasts HR dynamics using linear time-invariant (LTI) models, and examine whether these context-sensitive parameters serve as interpretable markers of athlete-or game-level stressors.
Methods: We used HR data from 72 university football players over two seasons. First-order LTI systems, with steady-state gain (SSG) and time constant (τ), modeled this data. A gradient boosting regressor predicted SSG and τ for upcoming quarters, using workload, head impact, and BMI as inputs. Model performance was assessed via R2, RMSE, and MAE, compared to population average LTI models. Associations of SSG and τ parameters with athlete- or game-level stressors were also analyzed.
Results: The dynamically tuned LTI model significantly outperformed the fixed-parameter baseline in HR prediction accuracy. Furthermore, predicted LTI parameters demonstrated meaningful associations with exertion levels, head impact exposure, and athlete BMI.
Conclusion: Context-aware LTI modeling offers both accurate HR forecasting and interpretable physiological insights during competition. This approach lays the groundwork for real-time athlete monitoring and autonomous decision- support systems in high-performance sports environments.