This pilot study demonstrated that an experimental framework combining a controlled sleep-deprivation protocol, repeated driving simulations, and machine-learning-based HRV analysis can reliably discriminate between 'Rested' and 'Tired' vigilance states, with all classifiers achieving average accuracy greater than 85%.
Key Findings
Results
All machine learning classifiers using HRV features achieved average accuracy greater than 85% in discriminating between Rested and Tired vigilance states using participant-independent Leave-One-Subject-Out Cross-Validation (LOSO-CV).
Participant-independent validation (LOSO-CV) was used to assess generalizability across subjects
All classifiers evaluated surpassed the 85% average accuracy threshold
The classification used HRV features extracted from ECG recordings during the first (Rested) and last (Tired) driving sessions of the 30-hour protocol
The supervised machine learning approach used experimentally defined vigilance states as labels without requiring additional ground truth measures
Results
Subjective drowsiness ratings and psychomotor vigilance test performance were significantly altered in the Tired state compared to the Rested state, confirming the validity of the two-state experimental framework.
Twelve healthy young adults participated in a 30-hour sleep-deprivation constant routine protocol
Driving simulations were conducted every two hours throughout the protocol
Data collection included electrocardiography (ECG), subjective ratings of drowsiness, psychomotor vigilance test (PVT), and driving performance
The Rested state corresponded to the first driving session and the Tired state to the last driving session of the protocol
Results
Variability in driving behavior, used as an indicator of impaired driving ability, increased significantly from the Rested to the Tired state.
Driving performance was measured during repeated simulations conducted every two hours over the 30-hour protocol
Increased driving behavior variability was identified as a key behavioral marker of sleep deprivation-related impairment
The driving simulation data supported the framework's ability to generate two physiologically and behaviorally distinct vigilance states
Results
Several HRV features showed significant differences between the Rested and Tired states, with effect sizes ranging from moderate to large.
HRV features were extracted from ECG recordings collected during driving simulations
Statistical analysis was performed on the HRV features prior to machine learning classification
Effect sizes for significant HRV metrics were characterized as ranging from moderate to large
HRV features served as physiological markers of vigilance state without requiring behavioral or subjective measures as ground truth for classification
Background
The controlled sleep-deprivation constant routine protocol successfully reduced major behavioral and environmental confounding factors that typically hinder vigilance characterization during driving.
The constant routine protocol was specifically designed to minimize confounding factors such as behavioral and environmental variation
Decreased vigilance associated with drowsiness is described as a major cause of motor vehicle crashes
Prior physiological characterization and detection of drowsiness are noted to be difficult due to behavioral and environmental confounding factors
The protocol lasted 30 hours and included driving simulations every two hours
Conclusions
The experimental framework generated two distinct vigilance states at both the physiological and behavioral levels, enabling reliable HRV data labeling for classification independent of additional ground truths.
The framework separated data into 'Rested' and 'Tired' states based on timing within the sleep-deprivation protocol rather than requiring concurrent ground truth measures
The approach enabled detection of decreased driving performance
Both physiological (HRV) and behavioral (driving variability, PVT, drowsiness ratings) measures confirmed the state separation
The authors note findings should be interpreted within the context of a pilot study with a small cohort of 12 healthy young adults
What This Means
This research suggests that changes in heart rhythm patterns (heart rate variability, or HRV) recorded during driving can reliably detect when a person has become dangerously drowsy due to sleep deprivation. In this study, 12 healthy young adults stayed awake for 30 hours while repeatedly completing driving simulations every two hours. Researchers compared data from the first driving session (when participants were well-rested) to the last session (when they were severely sleep-deprived), and found clear differences in heart rhythm patterns, reaction times, self-reported drowsiness, and how steadily participants drove.
Using these heart rhythm measurements and machine learning computer programs, researchers were able to correctly identify whether a person was in a rested or tired state more than 85% of the time — even when the programs were tested on participants they had never seen before. This is important because it suggests the approach could potentially generalize to new individuals rather than only working for the people it was trained on. Several specific heart rhythm metrics showed moderate to large differences between rested and tired states, meaning the physiological signal of drowsiness was strong enough to be reliably detected.
This research matters because drowsy driving is a significant cause of traffic crashes, yet it is notoriously difficult to detect objectively. This study suggests that wearable heart rate monitors combined with machine learning algorithms could potentially serve as a non-invasive way to monitor driver alertness. However, the researchers caution that these findings come from a small pilot study of only 12 healthy young adults under strictly controlled laboratory conditions, so further research with larger and more diverse groups in real-world driving conditions would be needed before such a system could be practically applied.
Check Your Own Numbers
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
Duverger J, Moser E, Boudreau P, Ouimet M, Boivin D, Saidi A. (2026). Tracking Vigilance While Driving: Pilot Study of Heart Rate Variability Classification Under a Controlled Sleep-Deprivation Protocol.. Sensors (Basel, Switzerland). https://doi.org/10.3390/s26196104