Interstate highway traffic flow classification based on hybrid and ensemble machine learning methods

Interstate highways form vital arteries in national transportation networks, accommodating large volumes of commuter, commercial, and freight traffic. These highways are facing many critical challenges that range from fluctuating traffic volumes to accidents, construction delays, and weather-related disruptions, all contributing to severe congestion, delays, and safety hazards. Accurate real-time traffic flow prediction helps address these issues by providing data that may inform traffic control measures, dynamic routing, and incident management systems to improve efficiency, safety, and reliability on these vital transportation corridors. In the United States, state-level departments of transportation are responsible for overseeing the operation of highways, bridges, public transit systems, and traffic management. These agencies not only maintain and improve the roadways themselves but also implement advanced technologies in collecting related datasets for enhancing traffic flow, safety, and efficiency. In this paper, an hourly traffic flow dataset collected by the Minnesota Department of Transportation is used. After preprocessing, the dataset is cross-validated by Logistic Regression (LR), Rotation Forest Classifier (RFC), and Extreme Gradient Boosting classifier (XGBC). Then, hybrid and ensemble classifications were performed using Greylag Goose and Dung Beetle optimization algorithms (GGO and DBO), along with the Stacking Classifier ensemble method. Among the base models, the results showed that RFC always outperformed its counterparts with a test Accuracy of 0.957, POD of 0.957, F1 Score of 0.957, and MCC of 0.942. The hybrid models demonstrated considerable performance improvements over the base models. Among the hybrid models, RFGO was the best hybrid-performing model with a test Accuracy of 0.986, a POD of 0.986, an F1 Score of 0.986, and an MCC of 0.981. The ensemble classifier, LXRGO, also proved to be one of the best models, achieving a test Accuracy of 0.944, a POD of 0.944, an F1 Score of 0.944, and an MCC of 0.925.