Developing machine learning forecast models using the data of the urban traffic corridor generated via a microsimulation program

With the rapid growth in private vehicle ownership in recent years, traffic problems have reached a serious level, especially in urban areas. The most significant of these problems are traffic delays and the consequent increase in travel time. These delays result in an increase in fuel consumption and greenhouse gas emissions, causing economic and environmental harm. The purpose of this study was to develop machine learning models based on synthetic data generated from simulation data. A traffic corridor comprising seven intersections in Erzincan's city center was selected for this study, where 27,648 scenarios were created using a combination of intersection types, signaling times, vehicle volumes, and lane changes, and the AIMSUN traffic simulation program was used to model and simulate the scenarios. A machine learning model was developed based on data obtained from simulations regarding total travel time, delay time and IEM CO2 parameters for each scenario. R2, MAE, MSE, and RMSE error metrics were used to compare the performance of Linear Regression, Support Vector Machine, Decision Tree, Random Forest, AdaBoost, Gradient Boosting, and XGBoost algorithms included in the models. Furthermore, three different scenarios selected from 27,648 scenarios were compared with one another and with the results of the simulations. According to error metrics, SVM and Linear Regression were selected as successful models for Delay Time, XGBoost for IEM CO2, and SVM and XGBoost for Total Travel Time. Based on the coefficient values analyzed in these models, it was determined that the intersection geometry and automobile traffic volume variables were of paramount importance for all three parameters. The results of this study suggested that successful machine learning models were possible to be developed through the use of synthetic data along with simulation data.