Traffic Modeling and Management, Innovative Data and Methods

In this dissertation, the main objective is to propose innovative data and methods to model and improve traffic operation and management in three directions: detecting and analyzing the occurrence of accidents in highways in real-time, improving the operation of reversible lanes to reduce congestion, and addressing the impact of connected and autonomous vehicle (CAV) on daily traffic. Therefore, first, an Extreme Gradient Boosting (XGBoost) model is trained to detect the occurrence of accidents and SHAP method is employed to analyze generated features from traffic, network, demographic, land use, and weather data sources. Second, to reduce congestion of highways, integration of real-time data-driven techniques and an offline statistical approach is proposed to optimize the operation of reversible express lanes. To this end, different delay indices are generated to measure and improve the performance of reversible lanes in passing the traffic more efficiently. Third, to address the impact of CAV on daily traffic, the current study presents data-driven techniques to relate changes in network traffic flows resulted by implementing CAV technology to traffic network and built-environment characteristics. Therefore, a comprehensive set of features representing network characteristics and urban structure patterns are generated and different machine learning techniques are used to characterize changes in Average Daily Traffic (ADT) under implementation of CAV scenario.