Exploring driver decision-making in lane-changing: A human factors approach using AI and naturalistic data

Abstract

Anticipating lane-change patterns represents a crucial dimension within the intricate framework of lane-change decision-making, exerting a profound influence on the fluidity of traffic dynamics and the overarching spectrum of road safety. Previous studies have mostly focused on fixed sections of highways, missing the changing and complex traffic patterns that drivers experience throughout the entire highway journey. This study explores the behavioural dimensions of lane-changing by leveraging an innovative data collection approach using cost-effective 3D LiDAR technology integrated into an instrumented vehicle platform. This system enables real-time, high-resolution data capture under diverse driving conditions, including nighttime and low-visibility scenarios. The study introduces the Expressway Drive Instrumented Vehicle (EDIV) Dataset, which captures naturalistic driving behaviour from 60 drivers over approximately 8,100 km on Indian expressways. Beyond the mere prediction of drivers’ lane-changing events, the research delves deeper into the intricate composition of lane transitions, employing a sophisticated repertoire of Machine Learning (ML) methodologies. Notably, the Extreme Gradient Boosting (XGBoost) technique emerges as the preeminent contender, showcasing better efficacy in accordance with classification metrics. Culminating in the application of elucidatory Artificial Intelligence (AI) paradigms, such as SHapley Additive exPlanations (SHAP) values, to interpret the intricacies of XGBoost-derived insights into driving behaviour. By integrating human factors research with data-driven methodologies, this study contributes to the development of safer and more behavioural informed traffic systems in mixed traffic environments. © 2025 Elsevier Ltd