Prediction of Traffic Counts Using Statistical and Neural Network Models

Abstract

This paper compared two different models for predicting traffic counts based on land use and demographic variables for the City of Calgary. Land use and demographic characteristics were used as independent vari ables at the Dissemination Area (DA) (small geographic unit having a population range of 400–700) level in the City of Calgary. Traffic count data from the City of Calgary were used as the dependent variable to devel op statistical and Neural Network models. Negative Binomial count statistical models (with log-link) were developed, as data were observed to be over-dispersed. Neural Network models were developed based on a mul tilayered, feed-forward, back-propagation design for supervised learning. The results indicate that Neural Network models ensured fewer errors than the statistical model. Overall, the Neural Network model yielded better results in estimating traffic count than the Negative Binomial Regression approach also considered in this study. The Neural Network model can be particularly suitable for its better predictive capability. However, the statistical model could be used for mathematical formulation or for developing a better understanding of the role of explanatory variables in estimating traffic count.