Traffic flow characteristics that lead to crashes on urban freeways are examined. Since these characteristics are observed prior to crash occurrence, they are referred to as "crash precursors." The objectives are (a) to explore factors contributing to changes in crash rate for individual vehicles traveling over an urban freeway and (b) to develop a probabilistic model relating significant crash precursors to changes in crash potential. The data used to examine crash precursors were extracted from 38 loop detector stations on a 10-km stretch of the Gardiner Expressway in Toronto for a 13-month period. An aggregate log-linear model was developed relating crash rates to the selected crash precursors observed upstream of the crash site. The results of this analysis suggest that the variation of speed and traffic density are statistically significant predictors of crash frequency after controlling for road geometry, weather, and time of day. With the model, crash potential can be established based on the precursors obtained from real-time traffic data.
Abstract