Uncovering Nonlinear Urban Road Geometric Thresholds Using Mars Traffic Modelling

Abstract

Urban traffic facilities in densely populated metropolitan areas often experience performance degradation due to the complex interactions between road geometric characteristics and driver behavior. Conventional traffic modeling approaches generally assume linear relationships, which may fail to capture critical threshold effects in heterogeneous traffic environments. This study aims to identify nonlinear relationships between urban road geometry, driver behavior, and traffic performance, as well as to determine critical threshold values that significantly influence operational efficiency at a grade-separated u-turn facility. The research employed a quantitative approach using microsimulation and Multivariate Adaptive Regression Splines (MARS) modeling. A total of 62 simulation scenarios were developed by varying geometric parameters, including turning radius, gradient, and weaving section length, along with behavioral variables such as motorcycle proportion and time headway. Traffic performance was evaluated using average delay and through-traffic indicators, while MARS was applied to detect nonlinear interactions and critical knot points. The results demonstrate that the MARS models achieved high predictive accuracy, with generalized R-squared (GRSq) values of 0.6687 for delay and 0.9231 for through traffic. Critical thresholds were identified at a turning radius of 12 meters, a gradient of 4%, and a weaving section length of 40 meters. Motorcycle proportion and time headway were also found to significantly affect traffic performance. These findings confirm that traffic performance is influenced by nonlinear interactions between geometric design and driver behavior. Therefore, MARS provides an effective analytical framework for identifying design-sensitive thresholds and supporting more adaptive, behavior-oriented urban road planning and traffic management strategies.