Despite extensive research on traffic safety, the complex, non-linear spatial discrepancy between risky driving and actual accidents remains a significant challenge to quantify within diverse urban contexts. This study investigates the non-linear relationship between grid-level risky driving patterns and traffic accident occurrence in Daejeon, Korea, examining how these associations vary across different urban contexts. Using data collected from July 2023 to June 2024, the analysis incorporates GPS-based risky driving indicators, including rapid acceleration, deceleration, and sudden maneuvers from general passenger vehicles, thereby overcoming the limitations of previous studies reliant on commercial vehicle data. By adopting an H3-based spatial grid system, the study classifies areas into four quadrants based on median values of risky behaviors and accident counts, further categorizing them into “Matched” and “Mismatched” types to identify spatial discrepancies. Furthermore, the Explainable Artificial Intelligence (XAI) technique is employed to integrate regional variables—including population density, land use, and transport infrastructure—to uncover the key drivers of accident risks. Providing a significant methodological improvement over traditional linear models, the findings demonstrate that identical driving behaviors can yield different safety outcomes depending on local environmental interactions. Specifically, while driver behavioral factors directly explain accident frequency in matched regions, accident risks in mismatched regions are more significantly shaped by spatial environmental factors, such as green spaces and commercial land use, which override direct behavioral impacts. This study provides a robust framework for developing data-driven, region-specific traffic intervention strategies, including context-aware advanced driver assistance systems (ADAS) and spatially tailored traffic calming, to enhance urban safety.
Yeom et al. (Fri,) studied this question.