Real-time rear-end conflict prediction on congested highways sections using trajectory data

Predicting rear-end conflicts in advance can avoid potential crashes and significantly improve road safety, especially in congested road sections. Many existing studies adopt macroscopic aggregated traffic flow state features and or environment features for rear-end conflicts prediction, which seems to overlook the impact of the temporal trends of various features during the conflict process on the outcomes. Thus, this paper uses microscopic trajectory data of front and rear vehicles for conflict prediction and explored the impact of trajectory changes trend on conflicts formation. A Gated Recurrent Unit (GRU) is employed to learn and encode conflict and non-conflict trajectory data and perform binary classification. The model has a 93 % recall and a 1.41 % false alarm rate. The Local Interpretable Model-agnostic Explanations (LIME) tool also explains the relationships between predicted conflict probability and input microscopic trajectory data. From the time analysis of the input trajectory using LIME, the following conclusions can be drawn. In congested road segments, when the speed of the leading vehicle is below 3 m/s and the speed of the following vehicle is above 4 m/s, it has a significant positive effect on the occurrence of conflicts. And some aggressive acceleration behaviors of drivers have the positive effect also. In addition, the reasons for conflicts among most vehicles are identical Because their feature distributions are similar. These findings can provide targeted insights for the management of ATM in congested road segments.