Real-Time Crash Risk Prediction Models and Transferability Analysis on Freeways

Abstract:

The paper aims to investigate the realtime crash risk based on the High Definition Monitoring System data on G15 Freeway in Nantong, China. Matched casecontrol method and parameter filtering method based on random forest were utilized to build SVM (support vector machine) models for the crashes on three subsegments respectively. Results show that the SVM models based on high definition data collected by High Definition Monitoring System show better performance than those in existing studies. The transferability research was also conducted to verify the transferability of the proposed SVMs and results indicate that the models can be transferred to a certain extent. They could be applied in realtime crash prediction process on road segments nearby after the calibration of the parameters in the models and the transferred models have relatively higher prediction accuracy.

Key words:freeway; real-time prediction; crash risk; support vector machines (SVM); transferability