Understanding and Modeling Drivers’ En-route Diversion Behavior During Congestion

Mohammad Shapouri, Louisiana State University and Agricultural and Mechanical College


In the field of transportation, traffic assignment models primarily have been used to forecast driver route preferences, translating their choices into traffic flow patterns across networks. These models are grounded in distinct behavioral theories and strive to explain how drivers navigate routes based on network features and personal tendencies. Using an aggregation approach, conventional traffic assignment models distribute demand among paths, considering utility and attractiveness. Despite their prevalent use in transportation planning and operations, the fundamental behavioral assumptions of these models have rarely been thoroughly explored. This gap is further compounded by their limited consideration of real-time adjustments and choices made by drivers specifically during congestion.

The goal of this study was to examine drivers' diversionary behavior during congestion and develop a predictive statistical en-route diversion model that includes influential factors in the decision-making process. Based on these goals, the primary objective of this research was to describe the decision-making process of individual commuters facing congestion and quantify the influential factors in this process. This was accomplished in two steps. Initially, a theoretical framework was proposed for the route choice decision-making process of individual drivers during congestion. Subsequently, 90 observations (including both diversionary and non-diversion events) with potential influential factors were created by analyzing an extensive dataset comprising GPS records of 20,255,016 vehicles in Sydney, Australia. Various statistical models were developed and tested based on this observation dataset. Ultimately, after performance analysis a logit model comprised of the main three influential variables was selected. This model showed a misclassification rate of 15.4% on the test set while providing a good level of interpretability.

Also, this research found that experienced delay per congestion distance, and remaining distance to destination positively influence drivers’ decision to divert during congestion. On the other hand, the expected increase in arrival time using a diversionary route in comparison with arrival time under normal conditions negatively impacted drivers’ decision to divert. The degree of consistency of variable importance level held true across various model types demonstrated through statistical significance, Mean Decrease Accuracy, and Mean Decrease in Gini. This observation indicated these three variables were the most influential factors in drivers’ decision-making process for diversion during congestion based on the observed trajectories.