Injury Prediction in Competitive Runners With Machine Learning.

Area of Science:

• Sports Medicine
• Data Science
• Biomechanical Engineering

Background:

• Injury prevention is crucial for athletic success.
• Predicting athletic injuries remains challenging.
• Novel technologies and data science offer potential insights.

Purpose of the Study:

• To predict running injuries using machine learning.
• To analyze detailed training logs for injury prediction.

Main Methods:

• Utilized a dataset of 74 high-level runners over 7 years.
• Applied two analytical approaches: daily training load time series (10 features) and weekly aggregate features (22 features).
• Incorporated objective GPS data and subjective training exertion/success metrics.

Main Results:

• Machine learning models (bagged XGBoost) achieved AUCs of 0.724 (daily) and 0.678 (weekly).
• The daily approach demonstrated a reasonably high probability of correct injury predictions.
• The model based on preceding days' training load was particularly effective.

Conclusions:

• Machine learning effectively predicts a significant portion of running injuries.
• Training load data, especially from preceding days, is key for accurate prediction.
• This approach can help tailor training programs and prevent injuries.