Using Machine Learning to Explain Recreational Running Injury: An Exploratory Classification Analysis

Area of Science:

• Sports Medicine
• Biomechanical Engineering
• Data Science

Background:

• Running is beneficial for health but carries an injury risk that is not fully understood.
• Wearable technology and machine learning (ML) offer promising tools to address this challenge.
• This study aimed to identify key factors, optimal ML models, and sample sizes for future research on running injuries.

Purpose of the Study:

• To identify critical factors predicting running-related injuries using ML.
• To determine the best-performing ML model for injury risk classification.
• To estimate the required sample size for future studies on running injury development.

Main Methods:

• An exploratory classification analysis was conducted using data from 88 recreational runners over 12 weeks.
• Heart rate, inertial measurement unit, and GPS data were collected from wristwatches.
• Seven ML models were evaluated using 10-fold cross-validation, with performance metrics including accuracy, precision, and recall. SHapley Additive exPlanations identified factor importance.

Main Results:

• The dataset comprised 4758 run instances from 80 participants.
• Four ML models (CatBoost, GMB, Light Gradient Boosting Machine, GXBoost) demonstrated high performance (>0.90 average).
• CatBoost achieved 96% accuracy in classifying injured runs, with critical power identified as the most significant predictor.

Conclusions:

• Boosting ML models, particularly CatBoost, are recommended for future studies on running injury development.
• Critical power should be included as a key factor in predictive models.
• Results should not be interpreted as causal, and model performance may be overestimated.