Editorial: Machine learning for operator fatigue detection and monitoring with wearable electronics

Fatigue is a multifaceted and pervasive issue that affects the wellbeing of operators and is prevalent across many domains, including but not limited to healthcare, aviation, and industrial applications. Fatigue can have physiological, neural, and psychological components. It can present as an acute, fluctuating, prolonged, or chronic condition, such as burnout. Fatigue often has serious consequences as it impairs decision-making, attention, and motor performance. The proliferation of wearable devices and advances in machine learning have opened new avenues for continuous, objective, and personalized fatigue monitoring. This Research Topic brings together contributions that highlight various aspects of operator fatigue detection and monitoring. These include the classification of mental workload (Yuan et al.), the evolution of brain networks in fatigue (Abdollahpour and Artan), environmental improvements for fatigue (Schneider et al.), the prediction of burnout risk (Dei Rossi et al.), and next-day fatigue (Aboagye et al.).

The diversity of these studies emphasizes that fatigue cannot be reduced to a single measurable quantity; rather, it is inherently multidimensional. Physiological markers, particularly heart rate variability (HRV), are prominently featured throughout this Research Topic. Aboagye et al. combined HRV with activity and sleep metrics to accurately predict fatigue in post-COVID populations. Their findings further support the importance of autonomic nervous system dynamics in fatigue assessment. Dei Rossi et al. used similar metrics to predict burnout risk. Yuan et al. used HRV features to classify the mental workload of pilots during different flight phases. Schneider et al. used the P300 component of Event-Related Potentials (ERPs) via electroencephalography (EEG) to measure cognitive workload, whereas Abdollahpour and Artan used brain hemodynamic measures obtained via functional Near-Infrared Spectroscopy (fNIRS) for fatigue staging.

The complexity and variability of fatigue-related data remain major challenges for traditional classification and regression frameworks. Dei Rossi et al. demonstrated that ordinal representations may be more appropriate than categorical or continuous representations for burnout prediction. They reframed burnout prediction as a pairwise ranking problem, which improved alignment with fatigue severity. Temporal modeling has also emerged as a promising research direction. Aboagye et al. used physiological data collected in a single day to predict fatigue levels on the following day. Similarly, Yuan et al. analyzed how dynamic workloads transition across task phases, while Abdollahpour and Artan modeled fatigue as an evolving brain network. The studies in this Research Topic also demonstrated the relevance of ensemble and tree-based methods such as XGBoost (Aboagye et al.; Yuan et al.). Alternative representations are also emerging, such as recurrent neural network architectures (Dei Rossi et al.) and graph-based brain connectivity models (Abdollahpour and Artan).

These articles also illustrated deployment in operational or real-world environments. Yuan et al. demonstrated the feasibility of continuous workload modeling in aviation, using HRV measurements to assess pilots’ cognitive state. Schneider et al. demonstrated improvements in efficiency and ergonomic outcomes associated with a smart surgical lighting system. Aboagye et al. was able to predict daily fatigue levels in individuals with post-COVID fatigue (pCF).

However, several challenges remain when it comes to fatigue monitoring. First, generalizability is a key limitation, as the majority of studies on fatigue rely on relatively small, domain-specific datasets that often fail to transfer across populations and operational settings (Aboagye et al.; Yuan et al). The lack of widely accepted ground truths compounds the generalizability problem because it complicates model training and evaluation. One possible reason for this is that fatigue manifests in multiple forms and arises from diverse physiological, cognitive, and psychological processes. For example, Dei Rossi et al. showed that higher-level constructs, such as emotional exhaustion, are not easily inferred from physiological signals alone. As fatigue is not a single condition, but rather can appear in different forms (e.g., mental fatigue, physiological fatigue …), and diverse temporal presentations (e.g., chronic vs. acute), the field would benefit from the development of a more comprehensive taxonomy of fatigue. In light of advances in deep learning and the proliferation of wearable devices, there is a clear need and opportunity for large and multimodal public datasets that cover the different forms and temporal presentations of the fatigue spectrum.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The authors acknowledge the use of Microsoft Copilot to assist with grammar and clarity during the preparation of this editorial.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited and reviewed by: Edward Sazonov, University of Alabama, United States