A supervised machine-learning pipeline that infers how many barges a tug/tow vessel is pushing on inland waterways using only Automatic Identification System (AIS) tracking data, which reports vessel position but not barge counts. Labels are created by matching traffic-camera observations to AIS records, and maneuverability features (speed over ground, rate of turn, vessel characteristics, interaction terms) are engineered. A two-stage design first classifies barge presence (AdaBoost, F1 = 0.932) and then predicts barge quantity (a Random Forest plus AdaBoost ensemble, F1 = 0.886), with Bayesian optimization for hyperparameter tuning, enabling real-time barge-count estimates for lock scheduling, congestion reduction, dredging, and freight planning.
A machine-learning approach to estimate barge presence and quantity on inland waterways from Automatic Identification System (AIS) vessel-tracking data. AIS reports the position of tug/tow boats but …