Predicting driver fatigue using HRV measures and machine learning

Driver fatigue is a critical factor in traffic accidents and is strongly influenced by sleep deprivation and prolonged driving. This study examines physiological and subjective fatigue responses by combining Heart Rate Variability (HRV) indicators with the Karolinska Sleepiness Scale (KSS) and Rating of Fatigue (ROF) in a controlled driving simulation. Forty participants completed four driving sessions under normal sleep and sleep-deprived conditions, with HRV recorded in each session. The results show that increased driving duration was associated with autonomic imbalance, as reflected in lower mean RR, mean HR, RMSSD, and HF, which were associated with increased fatigue levels. In contrast, sleep duration did not significantly affect most HRV indices, although it had robust effects on subjective fatigue measures (ROF and KSS). Significant correlations were observed between HRV parameters and ROF, indicating that physiological changes aligned with subjective perceptions. To complement the statistical analysis, Logistic Regression and ensemble learning models were applied to classify fatigue-related conditions. XGBoost achieved the highest performance, with an accuracy of 77%, and identified Mean RR, mean HR, and LF as the most influential predictors. These findings indicate that combining HRV metrics with machine learning improves fatigue pattern detection and offers strong potential for developing real-time fatigue monitoring systems. The study highlights the importance of integrating objective and subjective measures to evaluate driver fatigue, particularly in conditions of sleep restriction and extended driving duration.