AI-Powered Monitoring of the Acute: Chronic Workload Ratio: Interpretable Injury Risk Prediction in Soccer Players

Abstract

Background:

This study proposes a model for monitoring the acute:chronic workload ratio (ACWR).

Hypothesis:

Historical training data are able to predict a soccer player’s future ACWR.

Study Design:

Cross-sectional study.

Level of Evidence:

Level 3.

Methods:

We propose a timeseries model built upon a Transformer-based foundation model -Tabular Probabilistic Forecasting Network for Time Series—based on historical training data from soccer players, incorporating sensor data (such as Global Positioning System or accelerometers) and athletes’ subjective feedback. We leveraged prompt engineering and large language models to enhance the model’s predictive capability, extracting previous knowledge-based artificial features from the DeepSeek model.

Results:

Our model achieved an average mean absolute error of 0.119, an average mean squared error of 0.029, an average root mean square error of 0.149, and an average R² of 0.564 in ACWR prediction. In addition, in ACWR_RISK prediction, the model achieved an accuracy of 87.12%, precision of 85.91%, recall of 87.12%, and an F1 score of 85.27%.

Conclusion:

Extensive experimental results demonstrate that the model predicts the future injury risk of soccer players effectively, helping players regulate workload fluctuations and maintain their training state and injury risk within an optimal zone.

Clinical Relevance:

The proposed model provides a practical tool for monitoring and predicting athletes’ workload dynamics, enabling early identification of elevated injury risk associated with abnormal ACWR fluctuations.