Injury management in professional football remains a critical challenge for safeguarding player health and team performance. While existing systems primarily focus on injury prediction and return-to-play timelines, the post-injury performance trajectory of athletes remains an underexplored dimension. This study introduces InjuryBoostStackNet, a novel hybrid ensemble learning framework designed to predict post-injury performance drops in professional football players. The model integrates Random Forest, Light Gradient Boosting Machine (LightGBM), and Multi-Layer Perceptron (MLP) as base learners, coupled with a Transformer-based attention meta-learner. Using a retrospective, non-clinical, publicly available performance dataset, the framework applies dual-mode feature selection through Random Forest Gini importance and SHapley Additive exPlanations (SHAP) values. Evaluation was conducted using stratified 5-fold cross-validation across eight standard metrics. InjuryBoostStackNet achieved an accuracy of 97.95% and an area under the receiver operating characteristic curve (ROC-AUC) of 0.993. While these results demonstrate strong discriminative performance, they also reflect the feature distributions of the dataset used in this study. In addition, SHAP-based visualizations provide transparent insights into influential predictors. This approach establishes a methodological foundation for injury impact forecasting and offers an interpretable framework to support future data-driven decision-making in sports management.
Wang et al. (Mon,) studied this question.