Multimodal traffic flow analysis and congestion prediction on expressways: evaluating the importance of machine learning models and features for improving prediction accuracy in urban traffic management

Abstract The increasing use of cars and urbanization has fueled traffic jams, infrastructure degeneration, and highway safety issues in cities today. Transport management is significant in supporting city economies due to the fact that congestion translates into a considerable amount of economic waste, such as time wastage, excessive use of fuel, and pollution to the environment. The author uses sophisticated machine learning algorithms, such as the Random Forests, Gradient Boosting, Decision Trees, and Logistic Regression, to process multimodal traffic streams and forecast the mass U.S. interstate freeway traffic congestion. The RIME algorithm is used as an extension of more advanced models, and it is used to optimize an important parameter set to achieve the best predictive integrity and efficiency. The data is divided into 80% and 20% to train and test, respectively, to guarantee sound model analysis. The evaluation of model performance is done through the use of feature importance analysis, confusion matrices, and Receiver Operating Characteristic (ROC) curves. There is an analysis that GB and LR are the best predictors, and RF gives a compromise of accuracy and training time. Findings indicate that good traffic management requires the utilization of hybrid predictive models and optimization methods. The research gives a valuable insight to policymakers and planners to develop policies that will minimize congestion and help to make the cities more mobile.