Temporal Safety Analysis of Light-Vehicle Crash Severity on Rural Roads: A Hybrid Machine-Learning and Econometric Modeling Approach

This study examines the impact of time-of-week variations on traffic patterns and driver behavior on rural roads, highlighting its significance in crash fatality modeling and safety systems. Temporal instability often leads to heterogeneous driving behaviors in crash severity analysis, yet few studies have incorporated time-of-week interaction terms (TOWINT) with other exogenous variables. To address this gap, the study develops a hybrid econometric and machine learning framework combining Latent Segmentation-based discrete choice Logit models (LSL) to account for both unobserved and systematic heterogeneity. The models—Latent Segmentation-based Random Parameters Logit (LSRPL) and Latent Segmentation-based Generalized Ordered Logit (LSGOL)—are estimated and compared across four scenarios. The feature subset selection algorithm based on association rules (FEAST) was used to identify six significant TOWINT, which were integrated into the final model. The LSGOL model, with three latent segments (low, high, and critical severity), provided the best model fit, with weekend crashes emerging as the most influential segmentation attribute underlying unobserved crash severity heterogeneity. The study found that on weekdays, mountainous terrain and drivers aged 60 and older were more likely to experience fatal crashes, while the presence of a shoulder reduced injury severity. On weekends, rollover and roadside crashes had higher likelihoods, with roadside crashes occurring nearly twice as often as during weekdays. By utilizing a novel crash severity modeling, this provides actionable insights for time-specific, model-driven decision-making, including shoulder widening (3–3.5 m) for mitigating weekday PDO crashes. Furthermore, continuous roadside protection, clear zone provision, and lateral offset enhancements for reducing weekend high severity crashes, ultimately supporting improved traffic management and safer driver behavior on rural roads.