A Data-Driven Approach For Duration Evaluation Of Accident Impacts On Urban Intersection Traffic Flow

Accurate and rapid estimation and prediction of accident impacts are significant, as it assists road administration to alleviate traffic congestion and help the road users to make better travel decisions. In either case, information of how long an accident affects the traffic at nearby intersections is essential. Although there have been various approaches regarding prioraccident issues, only a few studies have been able to evaluate the impacts after an accident. With the emerging of new traffic sensor technologies, traffic data have exploded. This inspires us to rethink the accident impacts evaluation problem based more on intensive data. Accordingly, a practical data-driven method is proposed in this paper, whose functions are twofold: 1) to identify the flow characteristic of each intersection based on the processed data and then quantify the accident impacts through outlier detection and 2) to evaluate the duration of impacts by means of hazard-based model with heterogeneity. The procedure developed in this paper will be useful for capturing the accident impacts duration at or near urban intersections, as well as identifying the causal factors that affect these impacts, which includes the accident features, road environment and the temporal characteristics of nearby intersections. These findings could make some valuable conclusions for a better traffic management.