Efficient axle-load control is essential for managing overloaded freight vehicles and limiting delays at highway weigh stations. Static weighbridges provide accurate confirmatory measurements but require vehicles to stop, whereas weigh-in-motion (WIM) systems screen vehicles in motion and have potential to increase throughput. This study develops a conditional machine-learning benchmark for evaluating the operational efficiency of axle-load control at the Athi River (Mlolongo) weighbridge station in Kenya. Operational records collected by the Kenya National Highways Authority between April and July 2026 were analyzed, comprising 1,410,389 vehicle passages, 5,808 station-hour observations, and 117 complete daily observations. A gradient-boosted regression model was used to predict static-scale confirmatory throughput from WIM-screened demand, weekday, month, and time trend. Performance was evaluated using five-fold cross-validation against fixed-constant and single-ratio benchmarks. A station-hour model was subsequently developed using hour, weekday, month, direction, hourly demand, and time trend, with eight-fold cross-validation grouped by calendar date. The daily conditional model achieved an out-of-sample R² of 0.84 and mean absolute error (MAE) of 88 vehicles/day, compared with R² = 0.00 and MAE = 227 vehicles/day for the fixed-constant model and R² = 0.39 and MAE = 184 vehicles/day for the single-ratio model. The station-hour model achieved R² = 0.90 and MAE = 4.7 vehicles/hour, compared with R² = 0.71 and MAE = 8.3 vehicles/hour for the hour-by-direction baseline. These results show that operational context substantially improves efficiency benchmarking. The findings support the use of WIM as a high-throughput screening mechanism, with suspected overloaded vehicles subsequently referred for confirmatory static weighing.