Research on Predicting Athlete Overtraining Risk Based on Adaptive Gate-Controlled Tree Ensemble Network
Zekun Wang, Li Pengfei, Xuefeng Zhang · Applied and Computational Engineering · 2026
In the process of the scientific development of modern competitive sports, precise control of training load is the core key to improving athletes' competitive performance and avoiding sports injuries. Overtraining syndrome seriously affects athletes' careers. Traditional monitoring
Methods: are difficult to accurately predict risks. This study collected 2018 longitudinal monitoring samples of endurance athletes, covering basic information, training load, sleep status, and physiological and biochemical indicators, etc. Through correlation analysis, a heat map was drawn to reveal the correlation patterns among variables. To break through the limitations of the static weight fusion of traditional integrated algorithms, this paper proposes an adaptive gated tree ensemble network algorithm. This algorithm introduces a sample-level dynamic attention routing mechanism and a manifold feature projection perception module, combined with a soft voting regularization loss mechanism, to achieve dynamic weight allocation and high-order feature mining for base classifiers such as Random Forest, LightGBM, and XGBoost. Experimental
Results: show that the proposed algorithm has an accuracy rate of 84.90%, a precision rate of 75.40%, a recall rate of 70.66%, and an F1 score of 72.55%. All indicators are superior to traditional machine learning algorithms such as AdaBoost, Random Forest, and SVM, verifying the effectiveness and superiority of this
Method: in the task of predicting the risk of overtraining in athletes.