The likelihood of crash or sometimes referred to as “crash potential” is significantly affected by traffic flow conditions. In general, as traffic flow becomes more turbulent, drivers are required to adjust speeds more frequently and crash potential is likely to increase. Since traffic flow conditions vary over time and high turbulence of traffic flow can occur at any time, crash potential is also time-dependent and must be estimated in real-time. The real-time information on crash potential can be effectively used to prevent crash occurrence through driver warning systems that monitor current traffic condition and advise drivers of high crash potential. The driver warning systems influence on drivers' behavior, such as having them reduce speeds, so that overall crash potential can be reduced and a crash can be avoided. For the design, operation and evaluation of the driver warning systems for the purpose of improving traffic safety, we should identify the relationship between the variation of traffic flow conditions and crash potential.

So far, the most existing crash potential models have used static measure of traffic flow (e.g. average daily speed) for the prediction of crash potential. Since these models are not suitable for real-time application, we need to develop a model that links real-time traffic flow characteristics to the time-varying crash potential.

Therefore, the objectives of the thesis are 1) to identify factors contributing to changes in crash potential (defined as “crash precursors”) for individual vehicles traveling over an urban freeway using real-time traffic flow data;​ 2) to develop a probabilistic model relating significant crash precursors to changes in crash potential while controlling for other traffic environmental factors;​ 3) to develop a methodology for the evaluation of a real-time crash prevention strategy using the proposed model;​ and 4) to evaluate the potential impacts of using variable speed limits as the intervention strategy to reduce crash potential.

The study identified the three important precursors, namely the variation of speed at a fixed location, the speed difference between upstream and downstream locations, and the cross-sectional (between lanes) correlation of volume difference between upstream and downstream locations. The study also found that a log-linear crash prediction model could explain the effect of the three crash precursors and other traffic environmental factors on crash potential in real-time by appropriately adjusting exposure to the time-varying traffic conditions. Finally, the study demonstrated the potential safety benefits of one real-time control strategy (variable speed limits) in terms of reducing crash potential, using the proposed crash prediction model integrated with a microscopic traffic simulation model.