Effects of accompanying HMI warnings on lane-changing safety margins in foggy highway work zones

OBJECTIVE: Low visibility and lane closures in foggy highway work zones increase the risks of forced lane changes. However, the temporal buffer available after lane-change completion before reaching the work-zone closure remains insufficiently characterized. This study proposed a Lane-Changing Safety Margin (LCSM) metric and developed an accompanying human-machine interface (HMI) warning system. An interpretable predictive framework was further established to predict LCSM and identify the contributions of external environmental variables, individual-characteristic variables, and driver-state and simulation-perception variables. METHODS: A high-fidelity driving simulation experiment was conducted with 35 licensed drivers. Eight scenarios were designed by combining lighting conditions (daytime/nighttime), traffic density (800/1600 pcu/h), and the presence or absence of accompanying HMI warnings. Fine-grained driving behavior data were collected and used to calculate the LCSM. An XGBoost model was developed to predict the LCSM, and SHapley Additive exPlanations (SHAP) were used to assess global feature importance, variable main and interaction effects, and individual-level predictions. RESULTS: (0.8133), whereas the IEM yielded a slightly lower RMSE than the FVM (4.4173 vs. 4.4511). The relative contributions of external environmental variables, individual-characteristic variables, and driver-state and simulation-perception variables decreased in that order. Traffic density and accompanying HMI warnings were the two most influential predictors. High traffic density was associated with a lower LCSM, whereas accompanying HMI warnings consistently increased the predicted LCSM. The effect of driving experience varied across traffic-density conditions, whereas fatigue exhibited a nonlinear relationship with the LCSM. CONCLUSIONS: Accompanying HMI warnings were associated with larger LCSM values, which corresponded to earlier lane-change completion patterns and increased residual temporal margins during forced lane changes under low-visibility highway work-zone conditions. The proposed XGBoost-SHAP framework combines accurate LCSM prediction with interpretable analyses of key predictors and their interaction effects. This framework provides quantitative support for real-time prediction of lane-change safety margins, HMI warning strategy optimization, driver-specific interventions, and work-zone traffic management.