The Federal Highway Administration (FHWA) vehicle classification scheme is designed to serve various transportation operational and planning needs. Many transportation agencies rely on Weigh-In-Motion and automatic vehicle classification sites to collect vehicle classification count data. However, these systems are not widely deployed due to high installation and operations costs. One cost-effective approach investigated by researchers has been the use of single inductive loop sensors as an alternative to obtain FHWA vehicle classification data. However, most models do not accurately classify under-represented classes, even though many of these minority classes pose disproportionally adverse impacts on pavement infrastructure and the environment. As a consequence, previous models have not been able to adequately classify under-represented classes, and the overall performance of the models are often masked by excellent classification accuracy of the majority classes, such as passenger vehicles and five-axle tractor trailers. This project developed a bootstrap aggregating (bagging) deep neural network (DNN) model on a truck-focused dataset obtained from Truck Activity Monitoring System (TAMS) sites, which leverage existing inductive loop sensor infrastructure coupled with deployed inductive loop signature technology, and already deployed statewide at over ninety locations across all Caltrans Districts. The proposed method significantly improved the model performance on truck-related classes, especially minority classes such as Classes 7 and 11 which were overlooked in previous research studies. Remarkably, the proposed model is also capable of distinguishing classes with overlapping axle configuration, which is generally a challenge for axle-based sensor systems.