Behavioural mechanisms of driver compliance under lane-level variable speed limits: field evidence and predictive modeling

Lane-level variable speed limit (VSL) systems can improve traffic safety and efficiency, but their effectiveness depends on driver compliance. This study investigates behavioural mechanisms of compliance under lane-level VSL and develops a prediction framework combining Recursive Feature Elimination and Light Gradient Boosting Machine (LGBM), benchmarked against six models using real-world vehicle trajectory data. Compliance is most strongly associated with lane speed limits and vehicle type, with upstream speed and lane position secondary. Where both lanes share the same limit, lower limits are associated with substantially lower compliance, particularly in the left lane. Where limits differ, compliance depends on traffic conditions and the speed differential. Anticipatory deceleration near the first gantry and stable lane allocation are strongly associated with downstream compliance. LGBM achieved 93.5% accuracy at a fraction of the computational cost of XGBoost. The findings inform static lane-level VSL design accounting for vehicle-type heterogeneity and anticipatory behaviour.