Real-time freeway crash risk analysis considering unobserved heterogeneity: introduction of aggressive driving behavior data

Crash risk on expressways is influenced by interactions among aggressive driving, traffic flow, and road alignment, yet their heterogeneity remains insufficiently understood. This study uses Floating Car Data (FCD) to derive traffic flow and aggressive driving indicators and develops a Random Parameters Logit model with Heterogeneity in Means and Variances (RPL-HMV). Monte Carlo sensitivity analysis of sampling ratios from 1:1 to 1:7 identifies 1:7 as the most robust ratio for reducing sampling bias. Eight risk variables are identified, including lateral curvature, rainy conditions, speed variability coefficient, and emergency braking, with speed variability showing significant random effects. Road segment length and temperature drive mean and variance heterogeneity, respectively. These findings clarify the effects of traffic and driving characteristics on crash risk.