Modeling Delays at Signalized Intersections under Mixed Traffic Conditions

Abstract

Control delay is a key metric for evaluating traffic efficiency at signalized intersections, and its accurate estimation is crucial for effective intersection management and signal optimization. Traditional field studies are time consuming, and analytical models often underperform, particularly in oversaturated traffic conditions. The use of artificial intelligence (AI) techniques for delay estimation is gaining attention to address this issue. However, in developing countries, where heterogeneous traffic is prevalent, suitable AI-based models are scarce. This study introduces three novel AI techniques, namely multigene genetic programming (MGGP), gene expression programming (GEP), and functional network (FN), to fill this gap. Data from 20 signalized intersection approaches across four Indian cities were collected to train and test these models. Key predictors of control delay, including green ratio, percentage of vehicles arriving during the green phase, average queue length, and degree of saturation, were identified. Utilizing these variables, both MGGP and GEP demonstrated strong predictive capabilities, slightly outperforming FN. These models offer simpler regression-like structures, making them more practical for field applications. Sensitivity analyses of the models revealed that the average queue length has the greatest influence on delays, emphasizing the importance of quick queue dispersion to minimize intersection delays. The outcomes of this study would be beneficial for improving traffic management and mitigating delays at signalized intersections in developing countries, where managing heterogeneous traffic is a significant challenge.