Predictive modeling of gaze patterns in drivers: a machine learning approach with tobii glass 2

Understanding and predicting drivers' gaze patterns is essential for improving road safety and optimizing in-vehicle displays. This study delves into the nuanced dynamics of drivers’ visual attention across varied road segments, employing both statistical analyses and machine learning models. Ten participants, spanning diverse demographics, participated in a real driving experiment, navigating curves and straight stretches while their eye movements were tracked using Tobii Pro Glasses 2. Statistical analysis unveiled significant variations in gaze behavior, emphasizing specific Areas of Interest (AOIs) like the instrumental panel, left view, main view, and right view during curves. Machine learning models, including XGBoost (XGB), Adaboost, Support Vector Machine (SVM), and an Ensemble Model, were deployed to predict gaze patterns. Adaboost emerged as the top-performing model, showcasing robust accuracy (82.50%). The Ensemble Model, capitalizing on the strengths of individual models, demonstrated a well-balanced performance with a remarkable training accuracy of 99.36% and testing accuracy of 82.50%, coupled with an F1-Score of 83.72%. Despite participant-related limitations, this study provides indispensable insights into the intricate dynamics of driver gaze behavior. It underscores the effectiveness of machine learning in understanding and predicting drivers' gaze patterns, offering valuable implications for applications aimed at fortifying road safety measures and optimizing in-vehicle displays. The balanced performance of the Ensemble Model affirms the potential of amalgamating diverse models, presenting a promising avenue for future research and practical applications in the realm of driver behavior analysis.