Crash frequency analysis of different types of urban roadway segments using generalized additive model

IntroductionThis paper utilizes generalized additive model to explore the potential non-linear relationship between crash frequency and exposure on different types of urban roadway segments.MethodsGeneralized additive models are used to analyze crash frequency data and compared with the commonly used crash rate method and generalized linear models using a five-year crash data set from Houston, Texas.ResultsThe study shows that the relationship between crash frequency and exposure varies by segment type and the linearity may only approximately exist in certain segment types. In addition, the generalized additive modeling results suggest that such relationship curves may not be monotonic. Finally, this study demonstrates that generalized additive models in general provide better flexibility and modeling performance than generalized linear models.Impact on IndustryThe generalized additive model provides a very promising alternative for crash frequency modeling and other safety studies.