A novel Bayesian hierarchical model for road safety hotspot prediction

Area of Science:

• Road safety
• Statistical modeling
• Bayesian inference

Background:

• Road safety management often relies on reactive measures.
• Identifying future accident hotspots proactively is crucial for effective intervention.
• Existing models may not fully capture site-specific trends or multiple time periods.

Purpose of the Study:

• To develop a Bayesian hierarchical model for predicting future accident counts at potential road safety hotspots.
• To enable a proactive approach to road safety scheme implementation by identifying likely future hotspots.
• To rank sites based on their potential to exceed a threshold accident count.

Main Methods:

• Utilized a Bayesian hierarchical model incorporating accident counts from multiple time periods.
• Extended the empirical Bayes formulation to include historical data from various years.
• Employed the Bayesian posterior predictive distribution for predictions and uncertainty quantification.
• Incorporated site-specific variations and accounted for unobserved factors.

Main Results:

• The model accurately predicts future accident counts.
• Point estimates from the predictive distribution closely matched observed counts.
• Demonstrated the ability to incorporate multiple time-points, giving more weight to recent data.
• Showcased the capability to offset global trend estimates with local, site-specific accident variations.

Conclusions:

• The proposed Bayesian model accurately predicts future accident counts.
• The model supports proactive road safety management by identifying potential hotspots.
• It offers a robust framework for incorporating multiple historical data points and site-specific factors.