Prediction of instantaneous perceived effort during outdoor running using accelerometry and machine learning

Area of Science:

• Sports Science
• Biomechanical Engineering
• Wearable Technology

Background:

• The rate of perceived exertion (RPE) is a subjective measure for training load, but its accuracy can be limited.
• Objective biomechanical data may offer a more precise assessment of exercise intensity.

Purpose of the Study:

• To develop machine learning models for predicting RPE during running using smartwatch-derived biomechanical parameters.
• To evaluate the accuracy of subject-independent and subject-dependent prediction models.

Main Methods:

• Forty-three runners completed a 5-km race, reporting RPE periodically.
• Biomechanical data (heart rate, cadence, stride length, etc.) were collected using a Garmin 735XT smartwatch.
• Machine learning regression models were trained to predict RPE based on biomechanical data.

Main Results:

• Subject-independent models achieved an average RMSE of 1.8 RPE points (approx. 12% relative error).
• Subject-dependent models significantly improved prediction accuracy, with RMSE as low as 0.45 RPE points (approx. <7% relative error) using 20% of training data.
• All models tended to underestimate maximal RPE by approximately 1 point.

Conclusions:

• Biomechanical data from commercial smartwatches can effectively predict perceived exertion in runners.
• Subject-dependent models offer superior accuracy for personalized RPE prediction.
• This approach enhances objective monitoring of training workloads.