Tracking vigilance fluctuations in real-time: a sliding-window heart rate variability-based machine-learning approach

Area of Science:

• Physiology
• Machine Learning
• Cognitive Science

Background:

• Real-world vigilance evaluation using heart rate variability (HRV) is promising but limited by slow feature extraction and subjective benchmarks.
• Objective and efficient HRV-based vigilance assessment is needed for practical applications.

Purpose of the Study:

• To enhance objectivity and efficiency in HRV-based vigilance evaluation.
• To associate HRV and behavioral metrics using a sliding window approach for real-time analysis.

Main Methods:

• Forty-four healthy adults performed vigilance tasks under varied conditions with electrocardiogram recording.
• A 30-second sliding window with a 10-second step was used for HRV and behavior analysis.
• Machine learning classifiers (SVM, k-NN, AdaBoost, random forest) were trained and validated for vigilance classification.

Main Results:

• Vigilance performance showed instability, particularly after sleep deprivation, linked to decreased heart rate and increased HRV.
• Support Vector Machine (SVM) achieved 89% accuracy in binary classification (high vs. low vigilance).
• SVM maintained 84% precision in identifying low-vigilance epochs in three-class classification.

Conclusions:

• Sliding-window HRV metrics effectively capture vigilance fluctuations during tasks.
• This approach enables more timely and accurate detection of performance decrements.
• The study demonstrates the potential for improved real-time vigilance monitoring.