Modeling traveler behavior in managed lanes using large-scale real-world data

Understanding travel behavior is a key aspect for a variety of transportation projects that aim to improve mobility. One of the primary challenges in analyzing behavior in traffic is to estimate drivers' willingness-to-pay to reduce their average travel times or to experience more reliable trips. Majority of existing literature on estimating valuation of travel time and its reliability depend on stated preference studies. This thesis employs large size revealed preference data to explore the observed behavior of drivers using State Road 167 in Washington.;In this thesis schedule delays are considered as measures of unreliability. First, a time-allocation model-based methodology for the valuation of schedule delays is developed. An analytical approach is provided to determine the parameters to be included in the utility function associated with travel alternatives. Developed methodology relaxes the constant marginal utility assumption in the conventional models, which is regarded as an economical restriction and a non-linear utility function is obtained to estimate the valuation of schedule delays by calculating the ratio of the marginal utility of schedule delay to the marginal utility of travel cost.;Following chapters focus on utilization of Big Data from tolling records in State Road 167 for estimating willingness-to-pay measures. Data set is processed for implementing mixed logit and dynamic mixed logit model specifications. In order to address variation in time-of-day and different travel directions separate models are estimated for different cases. The results show that there exist considerable differences in valuation of schedule delays and valuation of travel times for the tested cases. This empirical finding supports the fact that there exists significant variation in drivers' willingness-to-pay which is mostly time- and travel direction-dependent and difficult to capture with limited number of observations from stated preference studies. Additional improvements are sought by randomly adding socio-demographic characteristics into the dynamic mixed logit model and it is seen that there are possible benefits of using hybrid data sets that include both revealed preference and socio-demographic variables.;Next, learning behavior in toll paying users in State Road 167 is investigated using Bayesian-Stochastic Learning Automata methodology. The findings showed that drivers are less reactive to the feedback from previous experiences compared to the earlier estimations for other tolled facilities. Finally, traffic simulation-based analysis is given to show a potential real-world implementation of the developed econometric models.;Recent developments in mobile networks, cloud computing and fellow technologies have led an exponential growth in data production and also increased the size of data that can be stored for later analysis. The abundance of publicly accessible data brings a number of potential opportunities for researchers and scientists to better understand and evaluate real-life problems. The findings in this thesis show the usefulness of Big Data for modeling travel behavior in managed lanes. For future studies Big Data-driven analyses can provide better guidance in informed decision making.