Identification Of High-risk Traffic Violation Types And Their Spatial-temporal Characteristics

The occurrence of traffic crash is often related to human factors.Previous research suggested that dangerous driving behavior was one of the most important contributing factors to traffic crash.Traffic violation is such a kind that could reflect dangerous driving behaviors and habits.Thus,it is often expected that traffic safety could be improved by lowering down traffic violation frequency.However,there are hundreds of traffic violation types.With limited resources,it is necessary and important to determine what types are highly correlated with crash risk and needs to be dealt with priority.In this research,we focused on exploring high-risk traffic violation behaviors by mining complex relationships among traffic violation,traffic crash and other contributing factors.Moreover,the spatial-temporal characteristics of those violation types were also examined so that traffic enforcement can be implemented more efficiently.The major contributions are as follow.The correspondence between traffic violation and traffic crash.Since there are hundreds of traffic violation types,it is necessary to extract those highly correlated with traffic crash risk so as to provide evidence for traffic enforcement.In this research,a non-linear dimension reduction method was employed to extract main features from both traffic violation and crash records.Then,a clustering analysis was conducted to further explore multi-multi correspondence between various traffic violation and crash types.In doing so,six traffic violation types were derived for further crash modeling task.Crash modeling based on complex data mining technique.In order to explore complex relationships among crash and various contributing factors,a Bayesian network approach was introduced.A number of features were extracted from original records,including driver,vehicle,roadway,and environmental factors.Based on the results,some violation types were found to elevate crash risk with the combination of other factors.As such,high-risk violation types and combinations were identified.Spatial-temporal analysis and prediction of high-risk violation combinations.First of all,a spatial-temporal analysis was conducted on the overall high-risk violations.Based on those,violation-prone locations and time periods were identified,which provides useful knowledge to traffic enforcement.Then,three typical high-risk violation combinations were selected for model training.Time series model and GRU model were both developed to capture their temporal characteristics and predict their trend.Based on the results,GRU models performed better than time series model,with reasonable prediction accuracy.The results of this research can help traffic management department better understand the relationship between traffic crash and traffic violations.It also provides valuable information for traffic enforcement and pro-active traffic safety management.