Quantifying training response in cycling based on cardiovascular drift using machine learning

Area of Science:

• Sports Science
• Exercise Physiology
• Biotechnology

Background:

• Aerobic fitness is crucial for endurance sports performance, but individual training responses vary.
• Coaches face challenges in objectively monitoring athlete improvement, fatigue, and overreaching.
• Technological advancements offer new methods for quantifying training responses.

Purpose of the Study:

• To develop and validate a machine learning (ML) method for measuring athlete response to training.
• To quantify aerobic fitness level using cardiovascular drift and aerobic decoupling data.
• To differentiate between training responders and non-responders in endurance athletes.

Main Methods:

• Twenty well-trained cyclists underwent monthly aerobic fitness tests over five months.
• Data on aerobic decoupling (power-to-heart rate ratio) and cardiovascular drift were collected during 60-minute rides at 75% functional threshold power.
• Machine learning models (Logistic Regression, Variational Gaussian Process, k-Nearest Neighbors) were employed to predict training response.

Main Results:

• A strong linear correlation was observed between cardiovascular drift and aerobic decoupling.
• All ML models demonstrated high predictive performance, with cross-validation accuracy between 0.87 and 0.9.
• The Variational Gaussian Process model achieved the highest classification accuracy (0.93) for predicting training response.

Conclusions:

• Cardiovascular drift and aerobic decoupling are reliable indicators of training response.
• Machine learning provides a promising tool for monitoring training adaptations and fatigue in endurance sports.
• ML-driven insights can support personalized training decisions for coaches and athletes.