A Real-hime Prediction Model for Rear-end Crash on Two-lane Freeway

The existing models for real-time crash prediction are difficult to be applied in the freeway management system that high-resolution traffic data cannot be collected.In this study, a real-time prediction model for rear-end crash is proposed based on the traffic data collected using a single detector.Based on the traffic data collected by ultrasonic detectors on Qiyang Freeway in Yangzhou, Jiangsu Province, China, the methods of matched case-control and binary logistic regression are used to develop a real-time prediction model for rear-end crash for a two-lane freeway.Three spatio-temporal matrixes, including a flow matrix, a speed matrix and an average space headway matrix, are extracted from the traffic data 5-20 minutes before crashes.Eigenvalues of matrixes are introduced to simplify the modeling process and avoid a strong correlation among the parameters.Results show that overall accuracy of this proposed model is 85.7%, and accuracy of prediction for crash rate is 33.3%, with a corresponding false alarm rate less than 2%.Thus the performance and effectiveness of this proposed model is verified.