People with spinal cord injury (SCI) show impaired thermoregulation during exercise, making skin temperature a noninvasive indicator. This study applies hybrid Extreme Learning Machine (ELM) models optimized with Ant Lion, Dragonfly, and Evolution Strategy algorithms to predict skin and core temperature dynamics during graded arm-crank exercise in 32 participants (16 SCI, 16 controls). The Dragonfly-optimized ELM achieved the highest accuracy (R² = 99.705, RMSE = 0.014) with no significant difference between predicted and measured core temperatures (p > 0.05). Feature-importance and SHAP analyses identified peak power output, group, and stage as dominant predictors, indicating reduced peripheral thermoregulatory variability in SCI.
Yang et al. (Wed,) studied this question.