This research aims at developing accident prediction models to improve freeway safety. Also by reducing incident related congestion on freeways, response and evacuation times would also be reduced in emergency situations. There are considerable amounts of data that are collected and stored for ITS applications. This data includes speed, volume and occupancy provided by loop detectors. Most of these variables are known to be related to accident occurrence and patterns. Previous work has shown the effect of speed variation and volume on traffic safety. In most of the previous work, average or historical speed and volume data have been used. This study looks at detailed real-time data and its relationship to accident occurrence with the objective of determining whether it is possible to predict the potential of accidents before they occur. The data from a 13.25 mile segment of Interstate 4 in Central Florida equipped with loop detectors has been used. Preliminary analysis of detailed real-time speed data showed changes in speed upstream of accidents. Substantial variation in speed before the accident (both space and time) are found significant when compared to cases that experienced no accidents. The matched case-control logistic regression has been adopted and showed that the 5-minute average occupancy observed at the upstream station during 5-10 minutes prior to the accident along with the 5-minute coefficient of variation in speed at the downstream station during the same time have been found to affect the accident occurrence most significantly. This paper proves that real-time freeway loop detector data could be used in predicting accident likelihood 5-10 minutes before they occur. Therefore, Advanced Traffic Management Centers could attempt to prevent accidents by disseminating warning messages or adopting Variable Speed Limit techniques. The practicality of using real-time freeway loop detector data in predicting accident likelihood and hence prevent accidents is addressed.


Civil and Environmental Engineering



URL: https://digitalcommons.calpoly.edu/cenv_fac/232