Evaluation of the Predictive Capabilities of Simulated Peak Hour conflict Based Crash Prediction Models

Road traffic crashes are one of the major causes of deaths worldwide. A safety prediction model is designed to estimate the safety of a road entity and in most cases these models link traffic volumes to crashes. A major problem with such models is that crashes are rare events and that crash statistics do not take into account everything that may have contributed to the crashes. The use of traffic conflicts to measure safety can overcome these problems as conflicts occur more frequently than crashes and can be easily estimated using micro simulation. For the purpose of this thesis, simulated peak hour conflict based crash prediction models are developed for 113 Toronto signalized intersections and their predictive capabilities are evaluated. The effects of a hypothetical left turn treatment on crashes and conflicts are also explored and compared to the study conducted by Srinivasan et al (2012). Lastly, the transferability of SSAM prediction models is evaluated to explore how well the models predict crashes for Toronto intersections.