Municipalities increasingly use GPS and IoT data to enhance freight analytics and inform policy. E-commerce activities, driven by connected vehicles, improve visibility yet complicate distinguishing e-commerce vehicles from other freight. Freight impacts the neighborhood's emissions, safety, noise, and traffic, underscoring the need to precisely identify e-commerce activities. Current traffic flow modeling and data lack granularity for neighborhood-level vehicle movements. As such, this study proposes a predictive framework to classify e-commerce vehicle trajectories using integrated datasets such as vehicle configuration, customer metadata, GPS trajectories, and duty cycle data. Capitalizing on Artificial Intelligence (AI), we first developed a large language model (LLM) to filter e-commerce fleets from non-e-commerce vehicles with similar configurations. Second, feature engineering on duty cycles and trajectories supports training machine learning models, including XGBoost, logistic regression, k-nearest neighbors (KNN), support vector machines (SVM), and random forests (RF). The proposed AI-driven model is tuned through cross-validation and oversampling to address class imbalance. Our AI model achieves nearly 80% precision, effectively distinguishing e-commerce vehicles with similar configurations. These results enable huge potential for municipal freight analytics across multiple domains.
