Review of data-driven lane-changing decision modeling for connected and automated vehicles

Lane changing represents a pivotal driving behavior for connected and automated vehicles (CAVs). This study presents a comprehensive review of the latest advancements in data-driven lane-changing decision (LCD) modeling for CAVs. The initial phase involved conducting a knowledge graph co-occurrence analysis on keywords pertinent to data-driven LCD models. Subsequently, the extant research was encapsulated from two distinct viewpoints. The first perspective centered on the data sources employed, detailing the widely used data types, their inherent characteristics, the predominant settings in which they are applied, and the scenarios for which they are suitable. The second perspective focused on the LCD modeling methodologies, examining the prevalent approaches and the methods used for validation and assessment. Building upon these insights, the paper identifies three promising research directions for the development of data-driven LCD models in CAVs. These include the necessity for a more inclusive dataset that captures the nuances of driver behavior and the dynamics of mixed traffic environments, the exploration of innovative data-driven techniques, and the establishment of a unified test set along with standardized testing criteria. The outcomes of this research are anticipated to significantly contribute to the crafting of more accurate and efficient LCD models for CAVs.