A data-driven framework for quantifying thermoregulatory dysfunction in hypobaric aerospace environments: Integrating dynamic feature weighting and physiological significance evaluation

Introduction: Accurate assessment of the impact of aerospace environments on human thermoregulation remains a major challenge for astronaut health monitoring systems. Traditional methods relying on static weighting coefficients often fail to capture the dynamic redistribution of heat flux across the body under varying atmospheric conditions. Methods: We developed an attention-based deep learning framework to analyze multi-site skin-temperature time series under normobaric (sea-level) and hypobaric (simulated 4500 m) conditions, validated in controlled climate-chamber experiments. Model robustness was assessed using leave-one-subject-out evaluation and benchmarking against alternative temporal baselines. Results: The proposed framework achieved an environmental classification accuracy of 86.96%, revealing distinctive temperature distribution patterns specific to each pressure condition. Notably, under normobaric conditions, attention weights concentrated on the head (14.78%) and thigh (14.81%), whereas hypobaric exposure shifted the model's focus toward the calf (16.02%) and forearm (14.15%). These findings align with known physiological adaptations to hypobaric hypoxia, demonstrating the model's interpretability and biological plausibility. Conclusions: By integrating attention-based feature weighting with physiological interpretation, this framework offers a dynamic and explainable approach to thermoregulation assessment. It holds promise for enhancing intelligent, interpretable health monitoring systems in aerospace medicine.