Road Network Accident Hotspots Identification And Evaluation Based On Improved Natural Nearest Neighborhood Clustering

Exact road network accident hotspots analysis results enable driver understanding where, why, what, and when automobile crashes occur, and is crucial to creating a safe driving environment. Therefore, hotspots analysis, which involves four phases: identification, ranking, profiling, and treatment, is deemed a key issue in traffic safety strategies. In this paper, a hotspots identification method, DTH3 N, is proposed, while a hotspots inter-characteristic analysis method based on PCA is given. Main contents of this paper are listed as follows:Firstly, basic methods of road network accident hotspots identification and evaluation are introduced. Concept of Natural Nearest Neighbor(3N) and corresponding Clustering Based on Natural Nearest Neighborhood(CB3N) method, which are employed as the basic of proposed hotspots identification method, is presented in detail. Meanwhile, advantages and limitations of classic factor for road accident hotspot evaluation, Composite Safety Performance Index(CSPI), is analysis. Then Principal component analysis(PCA), which be used in accident hotspots accident cause analysis, is described.After that, proposed DTH3 N method is presented in detail and corresponding parameters selection criteria is given. Based on accident data provided by STATs19 database, a series experiments are done by MATLAB tool. Experiment results show that proposed DTH3 N method presents a better performance on cluster division than existing methods.Then, based on CSPI value, two accident hotspots causes factors are defined and accident hotspots causes factors composite weight index based on PCA are proposed. Experiment results show that outputs of proposed accident hotspots cause analysis method could be used as supplementary parameters for hotspots evaluation.