Spatiotemporal urban traffic safety analytical framework by integrating nonparametric approaches

Area of Science:

• Urban planning and transportation safety
• Statistical modeling and machine learning applications

Background:

• Over 75% of the population resides in urban areas, necessitating safe and inclusive transportation environments.
• Mitigating accident risks and enhancing inclusivity in city traffic requires advanced analytical approaches.

Purpose of the Study:

• To develop a novel nonparametric modeling framework, the Mixed-Effect Tree Ensemble with a Gaussian Process (ME-GP), for city-wide traffic safety analysis.
• To provide traffic safety information accessible to all stakeholders, from end-users to decision-makers.

Main Methods:

• Utilized police-reported accident data from Seoul, South Korea.
• Developed the ME-GP framework integrating nonparametric modeling, tree ensembles, and Gaussian processes to predict accident risks at the road-segment level.
• Accounted for spatiotemporal heterogeneity and unobserved data complexities, capturing nonlinearities with Gaussian processes.

Main Results:

• The ME-GP model demonstrated a 15% improvement in predictive accuracy and lower variance compared to other nonparametric models.
• Demographics, traffic conditions, and road structure were identified as the most significant factors influencing accident risks.
• Elderly individuals exhibited a 20% higher accident risk than youth, while children had lower risks due to protective measures like school zones.

Conclusions:

• The ME-GP framework offers a robust and reliable method for analyzing urban traffic safety.
• Findings highlight the nonlinear and spatiotemporally heterogeneous relationships between determinant factors and accident risks.
• The study provides crucial insights for developing safer and more inclusive urban transportation networks.