Application of gradient boosting tree algorithm in quantitative evaluation and program adjustment of college students’ physical exercise outcomes

Traditional physical education assessments often rely on isolated performance metrics or single-time fitness tests, which fail to capture the dynamic interplay between multidimensional behavioral and physiological indicators such as training frequency, cardiorespiratory adaptation, and recovery efficiency. This limitation obstructs the provision of timely, personalized feedback for college students. To bridge this gap, the present study establishes a gradient boosting tree (GBT) framework that synthesizes multi-sourced data—including weekly exercise sessions, heart rate variability, body mass index, and sleep duration—into a continuous efficacy score. Based on a sample of 220 undergraduates tracked over eight weeks, this data-centric approach aims to transform empirical training adjustments into evidence-based decisions, thereby providing a more holistic and dynamic evaluation system for university sports pedagogy. The GBT model exhibited strong predictive validity (MSE = 0.82, R 2 = 0.87), with exercise frequency (importance = 0.32) and heart rate fluctuation (importance = 0.25) identified as the dominant influencing factors. These findings enable a quantitative, individualized feedback loop, allowing targeted adjustments—such as optimizing weekly session frequency and intensity—to facilitate evidence-based training modifications.