Machine Learning Improves Truck-to-Shipment Matching Using GPS Data

The Intelligent Truck Matching (ITM) 2.0, a novel machine learning system, tackles the challenge of connecting trucks with shipments in full truckload logistics when vehicle identifiers are either absent or damaged. A paper published on arXiv outlines this system, which treats the matching process as a probabilistic ranking issue. By utilizing Uber H3 hexagonal spatial indexing, it transforms GPS signals into route similarity features, integrating temporal data, and employs LightGBM gradient boosting with threshold-based post-processing. The methodology underwent evaluation through offline model selection (SVM, XGBoost, LightGBM), ablation studies, and production shadow testing, demonstrating notable advancements compared to rule-based methods. This initiative aims to facilitate real-time tracking and improve estimated time of arrival (ETA) predictions for shipments lacking visibility.