Cycling network risk assessment utilizing crash data and road characteristics

Abstract

Traffic congestion has led cities to pursue cycling as an alternative mode of transport. However, cycling ridership remains limited due to concerns over safety. This study looks for a correlation between road characteristics and cycling crash risk on a network-level analysis using data from Berlin, Germany. Support Vector Machine and CatBoost algorithm were used to model cycling crash risk from road characteristics, and model predictions were analyzed using Shapley Additive exPlanations. It finds that while road characteristics have some correlation with cycling accidents, the correlation is not strong enough to be the sole indicator for risk. Intersection complexity greatly increases cycling risk, while roads and cycling infrastructure shared with pedestrians decreases crash risk for cyclists.