Adaptive HRV analysis: Reinforcement learning-driven training load monitoring in sports science

Heart rate variability (HRV) is a widely used biomarker for assessing physiological stress, recovery, and training load in sports science. During exercise, the mechanical changes in various parts of the body, such as muscle contraction and relaxation, joint movement, and the dynamic response of the cardiovascular system, are closely related to HRV. However, traditional analysis methods face significant challenges in handling HRV’s nonlinear dynamics and noise sensitivity. These limitations reduce their effectiveness in complex sport scenarios. To address these limitations, this study proposes an innovative HRV feature extraction framework that integrates reinforcement learning (RL) with an attention-based Long Short-Term Memory (LSTM) network. The framework dynamically optimizes feature selection and weighting through RL. The integration of an attention mechanism enables the model to prioritize critical temporal segments, improving its ability to capture and interpret key physiological patterns. Additionally, the model combines time-domain, frequency-domain, biomechanical factors, and nonlinear features, providing a comprehensive and robust representation of HRV signals. The framework was validated on four publicly available datasets covering resting, exercise, stress, and recovery states. It achieved an average accuracy of 95.0% and an F1-score of 90.8%, outperforming state-of-the-art baselines by 2.7% to 3.4%. These results demonstrate the proposed method’s superior performance in stress detection, training load prediction, and recovery assessment, establishing it as a scalable and adaptive tool for HRV-based sports training monitoring and health management. The framework’s innovative design offers significant advancements in the analysis of complex HRV data, paving the way for intelligent and personalized applications in sports science and healthcare.