Which Method Is Better for Developing Freight Planning Models at Seaports—Neural Networks or Multiple Regression?

Abstract

Ports are the primary generators of freight traffic in the United States. Seaport operations will require operational and infrastructure changes to maintain the growth of international cargo operations. Truck and rail trip generation and modal split models will provide transportation planners and public agencies with valuable information for prioritizing funds for roadway upgrade projects and port infrastructure modifications. Two methods are presented for developing freight trip generation models—regression analysis and back-propagation neural networks. These models are applied in predicting the levels of cargo truck traffic moving inbound and outbound at seaports. For the Port of Miami, the back-propagation neural network model was more accurate than the regression analysis model. However, the neural network model requires a sizable database. A second application of the back-propagation neural networks approach developed a truck trip generation model and a truck-rail modal split model for the Port of Jacksonville. The primary factors affecting truck and rail volume were found to be the amount and direction of cargo vessel freight, commodity type (bulk, break bulk, or liquid bulk), and weekday of operation. In summary, the neural network model results were found significantly accurate for both Florida ports.