Deriving workload from driving behavior and psycho-physiology in work zones
通过驾驶模拟实验，结合驾驶行为、生理数据和问卷调查，利用集成学习模型以93.14%的准确率识别工作区驾驶的工作负荷水平，并分析其与行为和心理生理指标的关系。
Due to atypical driving scenarios, work zones may increase driving uncertainty, elevate behavioral deviations and intensify psychological workload, potentially pose risks and even lead to crashes. Existing studies predominantly focus on regular scenarios, traditional models and theories, which are not able to explain work zone driving behaviors and psycho-physiology due to different driving environment and complexity. This study aims to better understand the dynamic interaction among work zone driving behavior, driver psychology and physiology, so as to develop an interpretable workload representation framework. A driving simulator study with one baseline and three work zone scenarios was conducted, along with the collection of driving behavior, physiological data, and post-simulator survey data. Sixteen features were inputted to Optuna framework for Bayesian hyperparameter optimization, then a stacking ensemble learning model was built to identify workload levels, with a 93.14% accuracy, excessing other ten machine learning models including Light Gradient Boosting Machine (LightGBM) and Neural Networks. Workload correlations with driving behavior and psycho-physiology were analyzed and visualized through SHapley Additive exPlanation (SHAP) method. Significantly higher workload is characterized by greater lateral position shifts, more frequent brake pedal utilization, higher heart rate, lower heart rate variability, and more often changes in pupil diameter and gaze position. The results of this study reveal the dynamic relationships among driving behavior, driver psychology and physiology under different work zone setups, which could provide clearer insight for the design and optimization of driver assistance and autonomous driving systems.