Machine learning model for aberrant driving behaviour prediction using heart rate variability: a pilot study...

Area of Science:

• Transportation Safety
• Human Factors Engineering
• Machine Learning in Healthcare

Background:

• Aberrant driving behaviors (ADB), such as speeding and abrupt maneuvers, pose significant safety risks.
• Existing methods for detecting ADB often rely on vehicle-based data, with limited integration of physiological indicators.
• Heart rate variability (HRV) parameters offer a potential non-invasive measure of physiological stress and fatigue.

Purpose of the Study:

• To investigate the predictive capability of machine learning models using heart rate variability (HRV) parameters for aberrant driving behavior (ADB) occurrence.
• To explore the relationship between sleep quality, driver fatigue, and ADB in professional bus drivers.
• To evaluate the performance of various machine learning algorithms in predicting ADB based on physiological and contextual data.

Main Methods:

• Collected naturalistic driving data, including driving behaviors and physiological data (HRV), from 10 highway bus drivers over 4 days.
• Utilized a navigation mobile application and heart rate watch for data acquisition, supplemented by self-reported sleep data and external traffic/weather information.
• Applied five machine learning models: logistic regression, random forest, naive Bayes, support vector machine, and gated recurrent unit (GRU) to predict ADB events.

Main Results:

• Drivers exhibiting ADB frequently reported low sleep efficiency (≤80%) and higher scores on sleepiness and driver error/lapse questionnaires.
• Significant differences were observed in HRV parameters between baseline and pre-ADB event measurements.
• The Gated Recurrent Unit (GRU) model achieved the highest prediction accuracy for ADB, ranging from 81.16% to 84.22%.

Conclusions:

• Sleep deficit is a significant contributing factor to increased driver fatigue and the occurrence of ADB.
• HRV-based machine learning models demonstrate potential for predicting fatigue-related aberrant driving behaviors in bus drivers.
• This research highlights the utility of physiological monitoring for enhancing road safety through early detection of at-risk driving states.