Analysis Of Spatial Characteristics Of Road Traffic Crashes Based On Data Mining Techniques

Road traffic safety is a social issue that has received much attention.The occurrence of traffic crashes mainly includes four factors: human,vehicle,road and environment.The road and environment are closely related to the spatial characteristics of traffic crashes,but the literature has been lacking in data collection and research methods of the analysis of spatial characteristics of road traffic crashes.With the rapid development of data science,various types of open data related to urban space are emerging,and data mining tools have also seen a considerable improvement,which provides a good foundation for studying the spatial characteristics of traffic crashes.Based on the road traffic crash data of Shenzhen,China from 2014 to 2016,this paper uses a variety of data mining techniques to study the spatial characteristics of road traffic crashes and explores specific application scenarios.This paper is first focused on the spatial environment characteristics of individual crash.An crash severity prediction model is established based on spatial characteristics using logistic regression and random forest algorithm,and the main spatial characteristics influencing the severity are also identified.Then,considering the impact of the heterogeneity of the crash on the performance of the model,the crash data sets are classified according to type,region and time,and the analysis results of the models under different classifications are compared.The research shows that the predictive performance of the models based on the above two algorithms is similar,but the random forest algorithm is preferred in terms of the data preprocessing efficiency and the interpretability of the results of influencing factor analysis.The classified results show that the classification can significantly improve the performance of the prediction model and can find the differences of influencing factors between different categories of crash.This paper subsequently considers the spatial distribution characteristics of a group of crashes.The geocoding method is used to spatially locate the crashes in order to establish the spatial relationship between them.Then,two kinds of spatial data mining techniques,density analysis and cluster analysis,are used to identify the area of high frequency and high severity of road traffic crashes in Shenzhen.The research shows that the spatial distribution of traffic crashes in Shenzhen shows a significant regional difference.The area density of crash frequency in the downtown Shenzhen is higher than that in the suburban areas of Shenzhen,whereas in per unit-length road,the crash frequency in suburban areas is higher than that in downtown.The area density of crash severity in suburban areas is also higher than that in downtown.The spatial characteristics of the crashes obtained by density analysis and cluster analysis are basically consistent.The cluster analysis can provide more abundant spatial feature information than the density analysis,while the density analysis is superior in computational efficiency.This paper finally takes the constraint of urban road network into consideration in studying the spatial distribution characteristics of crashes.The road network data is collected from Internet and the geospatial information of roads of different levels is extracted.The crashes and the urban road network are mapped to the urban space grid according to their latitude and longitude position.The mathematical morphology algorithm is used to optimize the road network structure.The inter-cell distance of the grid considering the distance penalty of the main road is then calculated.Based on this distance and spectral clustering algorithm,the crash investigation area division considering the main road constraint is realized.The effectiveness of the above scheme is verified by comparison with the actual scheme currently used in business.