Modelling the influence of traffic infrastructure characteristics on e-scooter accidents in the city of Zurich

Abstract. The rapid emergence of shared electric scooter (e-scooter) services has posed new challenges to road safety over the last few years as a serious worldwide public concern. Previous studies have investigated e-scooter accidents from multiple perspectives. However, research gaps still exist in understanding the role of infrastructure-related factors in e-scooter accidents. This study aims to investigate and model the relationship between the characteristics of traffic infrastructure and the presence of e-scooter accidents, especially the presence of curbs and the complexity of street views. Curb extraction was first achieved by applying the Segment Anything Model to Google Street View images as a supplement to existing traffic infrastructure data. Variables related to curbs, the complexity of street views, and traffic transport were then constructed. With pseudo-absence points being generated, the influence of traffic infrastructure on the presence of accidents was analysed by applying logistic regression and Random Forest classification. Results show that bicycle traffic count, distance to road, and object complexity of the whole scene are the most important variables in the classification model, while besides these three, the presence of curbs, distance to pedestrian crossing, speed limit value, urban district, and road width are significantly correlated with the presence of e-scooter accidents.