Bayesian Inference for Enhanced Road Safety Analysis

Abstract

This thesis proposes innovative methods to analyse road traffic collision data with the aim of improving the evaluation of before/after safety schemes and predicting collision hotspots, thereby addressing significant research problems in the field of road safety. We integrate Bayesian inference, extreme value theory, and spatio temporal modelling to create robust and flexible models for these analyses. Specifically, the research problems include evaluating and demonstrating the lim itations of frequently utilised techniques in the assessment of road safety schemes, and the need for bespoke modelling formulations for atypical before/after studies. Key findings are introduced in the form of a new model to capture treatment ef fect for randomised trials, and the application of extreme value theory to conduct a traffic conflict-based before/after safety scheme evaluation. To account for spa tial correlation between neighbouring sites, Gaussian processes are included in the expressions for the location and scale parameters governing the generalised extreme value distribution. In terms of hotspot prediction, a Bayesian hierarchical model is proposed to segregate the seasonal and zonal effects in monthly collision data from multiple sites within fixed Traffic Administration Zones. Additionally, a spatio temporal model for collision rates is introduced that allows for serial dependence, seasonality, and correlation between rates at nearby zones. The key impact factors of this research are manifested in its practical applications to real-world data, including collision rate data from north Florida, USA; traffic conflict data from Vancouver, Canada; STATS19 data from the UK; and collision count data from Tyne and Wear, UK. These applications demonstrate the effective ness of the proposed approach in improving the evaluation of safety schemes and predicting collision hotspots, thereby offering insights that can guide stakeholders in making informed decisions on road safety interventions.