Real-time Secondary Crash Likelihood Prediction Excluding Post Primary Crash Features
A new hybrid ML framework identifies 91% of secondary crashes while cutting false alarms to just 0.20.
Deep Dive
Researchers Lei Han, Mohamed Abdel-Aty, and team developed a hybrid AI framework for real-time secondary crash prediction. It uses a dynamic spatiotemporal window to analyze traffic flow and environmental data, excluding unavailable post-crash details. An ensemble of six ML models achieves 91% accuracy with a 0.20 false alarm rate and a 0.952 AUC score. Traffic management systems can use it to proactively deploy resources and mitigate congestion.
Why It Matters
Enables proactive traffic management to prevent congestion cascades, potentially saving time, reducing emissions, and improving road safety.