Geographically weighted machine learning for modeling spatial heterogeneity in traffic crash frequency and determinants in US

• DOI: 10.1016/j.aap.2024.107528
• Publication Date: 2024-03-05T10:56:49Z
• Authors: Shuli Wang; Kun Gao; Lanfang Zhang; Bo Yu; Said M. Easa
• Keywords: Spatial heterogeneity

Spatial analyses of traffic crashes have drawn much interest due to the nature spatial dependence and heterogeneity in crash data. This study makes best Geographically Weighted Random Forest (GW-RF) model explore local associations between frequency various influencing factors US, including road network attributes, socio-economic characteristics, land use collected from multiple data sources. Special emphasis is put on modeling effects a factor different geographical areas data-driven way. The GW-RF outperforms global models (e.g. Forest) conventional geographically weighted regression, demonstrating superior predictive accuracy elucidating variations. reveals distinctions certain frequency. For example, importance intersection density varies significantly across regions, with high significance southern northeastern areas. Low-grade emerges as influential specific cities. findings highlight zones. Road factors, particularly density, exhibit universally, while socioeconomic variables demonstrate moderate effects. Interestingly, show relatively lower importance. outcomes could help allocate resources implement tailored interventions reduce likelihood crashes.