IDENTIFYING AND PREDICTING SECONDARY CRASHES USING REAL-TIME TRAFFIC DATA: A MULTI-STATE ANALYSIS

Abstract

Secondary crashes, which occur because of primary incidents, present serious safety hazards and contribute to traffic congestion. This study develops a data-driven framework to identify and predict secondary crash occurrence using large-scale real-time traffic and incident data from Maryland, Virginia, Tennessee, and Florida. A hybrid identification approach integrating spatiotemporal, network-based, and probe-speed reduction criteria is used to dynamically capture the impact area of primary incidents. The study examines key factors contributing to secondary crash occurrence using logistic regression models. In addition, machine learning models are used to predict the likelihood of secondary crash occurrence based on traffic, incident, environmental, and roadway characteristics. The results indicate that average volume is the most important feature in all four states.