Machine learning and traffic simulation for corridor-level optimization in Intelligent Transportation Systems: reducing travel times, delays, and carbon dioxide emissions

Increased individual vehicle ownership has led to longer travel times, increased delays, and higher CO2 emissions in traffic networks. This study aims to optimize intersection designs, lane counts, and signal timings within an Intelligent Transportation System (ITS)-based traffic corridor to reduce travel time, minimize delays and decrease instantaneous CO2 emissions. To achieve this, the current conditions of a seven-intersection corridor were simulated, resulting in 27,647 different scenarios with various combinations of vehicle volumes, intersection properties, lane configurations, and signalization. Incorporating real intersection vehicle count data and signal timings further enhanced the accuracy of the simulations, providing a more realistic representation of traffic flow and helping to refine the optimization of intersection designs, lane counts, and signal settings. In total, 27,648 scenarios, including the baseline scenario, were analyzed, resulting in three datasets that focus on delay time, instantaneous CO2 emissions, and total travel time. These datasets were modeled using various algorithms, including linear regression, support vector regression, elastic net, decision trees, random forests, AdaBoost, gradient boosting, XGBoost, and CatBoost. Among these models, CatBoost was found to be the most effective in explaining delay time, instantaneous CO2 emissions, and total travel time. Feature importance analysis using CatBoost revealed the optimal designs for reducing travel time, minimizing delays, and lowering emissions: modern roundabouts at the Dortyol and Old Police School intersections, and a grade-separation at the Terzibaba intersection. The findings demonstrate that integrating machine learning with traffic simulation can support ITS by improving traffic flow, reducing congestion, and lowering CO2 emissions.