| Popis: |
Most bicycle accidents are inextricably bound up with risky riding behaviors, which crossing the street illegally at unprotected mid-block locations is nothing to sneeze at. Compared with cyclists crossing the street at the crosswalk or intersections, there is a huge risk of accidents when they ignore or disobey road rules and across recklessly. Yet, the misbehavior of cyclists is an under-explored area in cyclist research due to the limited availability of detailed cycling data. This study creatively develops a GPS-based detection framework to capture risky street-crossing actions for the cyclists from large-scale bike sharing trajectory data. A data-driven modeling approach, based on structural topic modeling (STM), is developed to reveal the complexity and regularity of cyclists' habitual risky crossing behavior. Since objective built environment is one of the key factors associated with cycling, another goal of this paper is to apply a gradient boosting decision tree (GBDT) model to disentangle how the features of built environment may influence the frequency of risky crossing events. The case study results show that risky street-crossing behavior is prevalent in bicycle traffic – for example, 16.94% of cycling trips are involved in illegal crossing action. Most cyclists engage in illegal crossing behavior at the approximate central part of the streets and during the day, which reveals the presence of heterogeneity over space and time. Strong correlations between commuting activities and risky street-crossing behaviors are identified from topic modeling. Meanwhile, the latent illegal crossing patterns unraveled here highlight that typical reasons for committing the risky riding action include the lure of the travel destination across the road and the inconvenience of riding round in distant legal crossing facilities. GBDT findings provide new insights on the existence of the association between built environment and cyclists' illegal crossing action. The places related employment and catering play a dominant role in contributing risky street-crossing behavior, and the influences of road length, road level, bus stop and metro station are not neglectable. Most built environment attributes show nonlinear correlations with crossing frequency. It is anticipated that this study would successfully shed a first light on the pattern of cyclists' risky street-crossing behavior at the metropolitan scale, and compliment engineering practices to improve crossing behaviors and bicycle safety. ISSN:0966-6923 ISSN:1873-1236 |
Full Text from ScienceDirect