1666324 - Proactive Traffic Management through Machine Learning based Predictions and Site-specific Insights

This study presents an approach for proactive traffic management that uses Machine Learning (ML) algorithms to predict imminent traffic congestion. To facilitate the proactive implementation of conventionally recommended traffic management strategies, the prediction of impending traffic congestion is formulated as a binary classification problem. Under this formulation, traffic data comprising of variables representing the lane dynamics and spatiotemporal state during uncongested conditions is labeled as belonging to one of two classes: free-flow or pre-congestion. The proposed formulation is tested using field data from two different freeway sites within the Atlanta metro area, and the underlying patterns present were learned through a set of classical ML classifiers. Additionally, the work also explores the influence of site characteristics on the applicability and transferability of the proposed approach for traffic congestion prediction. The tested sites consistently experience evening peak hour congestion, however, they differ in terms of the location of the most frequent active bottleneck. This difference has been demonstrated to impact the frequency of false positives and precision scores for each site. This inference has important implications for future application of ML algorithms for congestion prediction by suggesting that the underlying mechanism for congestion generation (e.g., bottleneck activation or congestion propagation) must be considered in ML algorithm training. The ML methodology developed in this study can be pivotal in advancing proactive traffic management strategies that can delay or minimize the onset of recurrent congestion.