This is like having a smart coach that not only predicts which players are likely to get injured soon, but also clearly explains why it thinks so, factor by factor, instead of being a mysterious black box.
Traditional injury risk models are either too simple or too opaque. Teams struggle to turn GPS, training load, and medical data into actionable, trusted insights about who is at risk and what to adjust. This work uses interpretable machine learning with SHAP to predict injury risk while giving clear explanations of the drivers behind each prediction.
Domain-specific feature engineering on rich athlete monitoring data (e.g., GPS, training load, wellness, medical history) combined with interpretable ML (SHAP profiles and thresholds) can become a proprietary knowledge asset and workflow embedded into a club’s or league’s high-performance ecosystem.
Classical-ML (Scikit/XGBoost)
Feature Store
Medium (Integration logic)
Data quality and standardization across teams and seasons; model recalibration as training methods, monitoring hardware, and game intensity change over time.
Early Adopters
Unlike generic sports analytics or black-box injury prediction models, this approach emphasizes model explainability via SHAP, making individual risk scores decomposable into contributions from specific load, wellness, or biomechanical variables—critical for practitioner trust and day-to-day decision-making in elite sport.