Abstract
Transportation agencies tasked with forecasting freight movements, creating and evaluating policy to mitigate transportation impacts on infrastructure and air quality, and furnishing the data necessary for performance driven investment depend on quality, detailed, and ubiquitous vehicle data. Unfortunately, commercial vehicle data is either missing or expensive to obtain from current data resources. Leveraging existing infrastructure, Hernandez et al. (8) developed a novel, readily implementable approach of integrating two exceptionally complementary data collection devices, Weigh-in-Motion (WIM) systems and advanced inductive loop detectors (ILD), to produce high resolution truck data. For each vehicle traversing a WIM site, an inductive signature was collected along with WIM measurements such as axle spacing and weight. As a case study, the researchers derived truck body configuration from this combined data source. Since body configuration can be linked to commodity carried, drive and duty cycle, and other distinct operating characteristics, body class data is undeniably useful for freight planning and air quality monitoring. Several significant improvements to the body classification model are made in this paper. First, a multiple classifier systems (MCS) method was adopted to increase the classification accuracy for minority body classes. Second, the model was expanded to all truck classes in the axle-based FHWA classification scheme. In all, eight separate body classifications models were developed from an extensive data set of 18,967 truck records distinguishing an unprecedented total of 23 single unit truck and 31 single and semi-trailer body configurations, each with over 80% correct classification rates (CCR). Remarkably, the body class model for five axle semi-tractor trailers â?? the most diverse truck category â??achieves CCRs above 85% for several industry specific classes including refrigerated and non-refrigerated intermodal containers, livestock, and logging trailers.