Modeling Distracted Driving: Analyzing Driver Gaze, Vehicle Positioning, and Psychological Response for Enhanced Traffic Safety

Driver distraction, resulting from attention shifts due to external factors or the use of in-vehicle devices, significantly heightens collision risks and varies in severity across different vehicle types, such as passenger cars and SUVs. Factors like alcohol impairment exacerbate these risks by affecting driver perception and reaction times, further compounding errors related to distraction while driving. This study analyzes data collected from 26 drivers using a multimodal Virtual Reality (VR) simulator, which captures biometrics, vehicle dynamics, and responses to distraction scenarios illustrated by varying grid sizes. Exploratory data analysis shows changes in psychological measures, such as Galvanic Skin Response (GSR), and vehicle positioning as distraction levels increase. Machine learning models (e.g., Multinomial Logit, SVM, Random Forest, XGBoost) and deep learning models (e.g., TabNet, RNN) are employed to evaluate the classification of driver distraction. Models like XGBoost and TabNet show superior performance, with insights into feature importance (e.g., Brake, Openness) improving interpretability. These findings provide essential directions for reducing driver distraction and enhancing traffic safety.