Estimating the effect of congestion on traffic safety risks: a double machine learning causal inference approach

Congestion severely threatens road traffic safety, but traditional accident data-based evaluation methods suffer from confounding variables and selection bias, making causal effect quantification difficult. Taking Sichuan Expressway as an example, this study proposes a causal framework integrated with double machine learning (DML) to accurately analyze congestion’s risk impact. It constructs a Traffic Order Index (TOI) for real-time dynamic risk characterization, applies the Conditional Shapley Value Index (CSVI) to screen confounding variables, and uses the DML model to mitigate estimation bias. The framework and risk assessment results can provide a scientific basis for expressway active safety prevention.