Spatiotemporal instability analysis of active traveller injury severities with small sample size and imbalanced crash...

Active traveller (including pedestrians and bicyclists) crashes pose significant challenges to sustainable transportation. Active traveller injury severities not only demonstrate temporal variations, but also differ across different functional zones within the city. Therefore, conducting a spatiotemporal analysis to understand the impact of various factors on active traveller injury severities can help develop effective strategies aimed at mitigating these severities. However, most existing studies mainly focus on temporal instability from year to year, ignoring the spatial difference between rural and urban areas. To examine spatiotemporal instability, this study uses North Carolina as a case study and divides the six-year (2017-2022) active traveller crashes into four sub-datasets according to distinct spatial and temporal characteristics. An explainable and balanced machine learning framework is designed to address the challenges associated with small sample size and imbalanced crash data and explore factors affecting active traveller injury severities. Results demonstrate that spatial instability has a greater impact than temporal instability. For instance, non-intersection, bicycle and travel lanes, medium speed limit and dark with light conditions are important in urban areas, but crosswalk areas are significant in rural areas. These results can help policymakers develop region-specific countermeasures to promote the reliability of active transportation systems.