Prediction of Perceived Exertion Ratings in National Level Soccer Players Using Wearable Sensor Data and Machine...

Area of Science:

• Sports Science
• Machine Learning in Sports
• Physiology

Background:

• Subjective ratings of perceived exertion (RPE) are crucial for monitoring training load in athletes.
• Understanding the relationship between objective load parameters and RPE is vital for optimizing performance and preventing overtraining.
• Elite soccer players present unique physiological demands that necessitate precise load management strategies.

Purpose of the Study:

• To identify key external and internal load parameters correlated with RPE in highly trained soccer players.
• To evaluate the predictive accuracy of various machine learning models for RPE.
• To develop and assess a deep learning architecture for RPE prediction in this population.

Main Methods:

• Utilized a large dataset (5402 training sessions, 732 matches) from 26 professional male soccer players.
• Collected 174 parameters including heart rate, GPS, accelerometer data, and RPE (Borg's 0-10 scale).
• Employed nine machine learning algorithms and one deep learning architecture, with rigorous data preprocessing and 5-fold cross-validation.

Main Results:

• The deep learning model achieved the highest predictive power for RPE (Mean Absolute Error: 1.08 ± 0.07).
• Tree-based models, particularly ExtraTree, showed high accuracy (MAE: 1.15 ± 0.03) and robustness to outliers.
• Maximal heart rate, maximal acceleration, and distance covered in a specific speed zone were the strongest predictors of RPE.

Conclusions:

• RPE prediction in elite soccer players is influenced by a combination of numerous external and internal load parameters, not a single variable.
• Maximal heart rate exerts the most significant influence on RPE.
• The ExtraTree model offers a robust, accurate, and broadly applicable tool for RPE prediction in soccer, runnable on standard computing platforms.