This paper aims to critically examine the validity of wrist-worn photoplethysmography (PPG) sensors in classifying exercise intensity zones in real time, with particular emphasis on the interplay between physiological, biomechanical and technological factors. This review addresses limitations in validity across different intensity domains and explores solutions integrating advanced analytics and signal processing.
Over 100 peer-reviewed sources were systematically reviewed to assess the validity and limitations of PPG sensors across treadmill, cycling and free-living conditions. The review dissects validity by intensity zones, explores inter-subject variability linked to skin pigmentation, BMI and age and evaluates algorithmic approaches from simple averaging filters to convolutional and recurrent neural networks.
Validity of PPG sensors significantly declines with increasing exercise intensity, primarily due to motion artefacts, vasoconstriction and low signal-to-noise ratios. Variability across devices and user demographics remains a major challenge. Studies show that hybrid sensor configurations (e.g. IMU + PPG) and machine learning-based artefact rejection (CNN–LSTM) enhance signal fidelity. However, most commercial wearables still rely on proprietary heuristics, limiting transparency and reproducibility. Cross-validation against ECG gold standards reveals that mean absolute percentage error increases up to 18% in high-intensity domains, raising concerns for zone-based training prescriptions.
This review bridges engineering, physiology and data science by offering a comprehensive synthesis of the mechanisms, limitations and solutions associated with PPG-based HR sensing in sports. It critiques the sufficiency of existing validation frameworks and advocates for standardised benchmarking, federated data sets and interpretable AI to guide future innovation.
