Predicting managed lane choice by integrating real world data and behavioral science

• Developed a multi-task learning (MTL) that jointly categorizes participants and predicts lane choices. • Benchmarked against ICLV, which incorporates latent traits and measurement error. • Travel time savings and variability are key factors of managed lane choice. • Psychological traits (e.g., NFCC, risk aversion) influence chooser vs. non-chooser behavior. • MTL model outperforms ICLV by capturing complex, non-linear relationships. Managed lanes (MLs) are a congestion management strategy that provides travelers with an option to either pay for uncongested travel on MLs or travel toll-free on adjacent general-purpose lanes (GPLs). Traditional travel demand models assume all travelers choose between these options for every trip based on utility maximization. However, recent research shows many travelers do not make such decisions trip by trip but instead rely on a habitual set of lanes. This study integrates real-world travel data with behavioral science to predict ML usage. A multi-task learning (MTL) model was developed to capture lane choice at two levels. At the upper level, the MTL model classified participants as choosers (who actively switch between MLs and GPLs) or non-choosers (who consistently use one type of lane). At the second level, the model predicted lane choice (MLs vs. GPLs), with each trip weighted by the participant’s probability of being a chooser. Data included detailed trip records from GPS logs and traveler information. The MTL model was compared with two models: an Integrated Choice and Latent Variable (ICLV) model and a two stage Random Forest classification model. The study analyzed data from 106 participants tracked over three months in Dallas-Fort Worth, Texas, and Northern Virginia. Results show that travel time variability, toll prices, and psychological traits (e.g., need for cognitive closure (NFCC)) significantly influence both the likelihood of being a chooser and the decision to use MLs. Participants with lower NFCC scores were more often choosers, consistent with expectations that those more comfortable with uncertainty are more likely to evaluate travel alternatives. While the observed behavioral patterns from the models align with established theory, the integration of psychological constructs such as need for cognitive closure (NFCC) and conscientiousness etc. with revealed-preference travel data contributes novel empirical insight into ML choice research. All models showed strong accuracy in identifying choosers and non-choosers, however, the MTL and ICLV hierarchical models outperformed the two-stage Random Forest model in predicting lane choice. While these findings provide valuable insights, a larger validation study in partnership with a managed lane operator is recommended.