Predicting Change in Average Vehicle Ridership on the Basis of Employer Trip Reduction Plans

Abstract

Artificial neural network (ANN) models are described, and efforts to build a model to predict changes in average vehicle ridership using about 7,000 employer trip reduction plans from three cities are highlighted. The development of the application is summarized; the neural network model performance is compared with other analytical approaches; and the results of the field test are summarized. Researchers at the Center for Urban Transportation Research combined the three data sets, identified model inputs and outputs from the data, and built the neural network model. This step also included building alternative models using regression and discriminant analysis to measure relative ANN performance. These models were compared with the FHWA’s transportation demand management model. The ANN model built only with data from Los Angeles was validated using a separate data set and evaluated according to the model’s ability to classify the change in average vehicle ridership (AVR) within an acceptable range. The final step was the validation of the model using data from other sites. The result was a model and software built on data from Los Angeles and Tucson that performed well when tested with data from Phoenix. On the basis of this project, the ANN model predicted an acceptable range of changes in AVR and was proven to be transferable to another city. Furthermore, the ANN model outperformed other analysis tools and was easier to use. Finally, the model provides a basis for helping to assess the impacts of employer trip reduction programs with minimal data collection requirements.