Causal Inference and Uncertainty Quantification for Counterfactual Analysis of Injury Severity in Multi-Vehicle Collisions

Document Type

Conference Paper

Publication Date

11-2025

Publication Title

2025 IEEE International Conference on Future Machine Learning and Data Science (FMLDS)

Abstract

Road crashes continue to pose significant public health concerns, prompting transportation agencies to prioritize strategies for improving roadway safety and mitigating crash impacts. Traditional approaches to modeling highway safety data often rely on regression-based methods to identify significant factors associated with injury severity. However, these methods primarily capture associations rather than causal effects, limiting their ability to evaluate the effectiveness of interventions. This study introduces a Bayesian causal inference framework to advance safety analysis by capturing underlying causal relationships on crash injuries. Using data from Ohio interstates, the framework assesses the effects of roadway, environmental, and behavioral factors on injury severity, with crash type modeled as a mediating variable. The average causal impacts and their associated uncertainties are estimated using the highest density interval (HDI). The study findings reveal that alcohol involvement and multi-vehicle crashes are key contributors to the increased risk of injuries and fatalities. Road geometry and vehicle speed exhibit both direct and mediated effects through crash type, while lighting conditions and the presence of work zones show uncertain impacts based on HDI. The results offer valuable insights for transportation safety policy, emphasizing the importance of targeting both direct and indirect pathways in crash injury prevention strategies.

Comments

Paper presented at 2025 IEEE International Conference on Future Machine Learning and Data Science (FMLDS) held in Los Angeles, CA, November 2-5, 2025

DOI

10.1109/FMLDS67896.2025.00077

Link to Full Text

Share

COinS