A New Spatial Count Data Model with Bayesian Additive Regression Trees for Accident Hot Spot Identification

TL;DR

This paper introduces a novel Bayesian spatial count data model utilizing Bayesian additive regression trees to improve accident hot spot identification by automating link function specification and capturing spatial correlations.

Contribution

It proposes a new spatial negative binomial model with Bayesian additive regression trees and Polya-Gamma augmentation, enhancing flexibility and uncertainty quantification in accident hotspot modeling.

Findings

Model performs as well as existing methods in goodness of fit.

Automates link function selection using machine learning.

Effectively captures spatial correlations and unobserved heterogeneity.

Abstract

The identification of accident hot spots is a central task of road safety management. Bayesian count data models have emerged as the workhorse method for producing probabilistic rankings of hazardous sites in road networks. Typically, these methods assume simple linear link function specifications, which, however, limit the predictive power of a model. Furthermore, extensive specification searches are precluded by complex model structures arising from the need to account for unobserved heterogeneity and spatial correlations. Modern machine learning (ML) methods offer ways to automate the specification of the link function. However, these methods do not capture estimation uncertainty, and it is also difficult to incorporate spatial correlations. In light of these gaps in the literature, this paper proposes a new spatial negative binomial model, which uses Bayesian additive regression trees to endogenously select the specification of the link function. Posterior inference in the proposed model is made feasible with the help of the Polya-Gamma data augmentation technique. We test the performance of this new model on a crash count data set from a metropolitan highway network. The empirical results show that the proposed model performs at least as well as a baseline spatial count data model with random parameters in terms of goodness of fit and site ranking ability.