Average and Peak Traffic Volumes: Neural Nets, Regression, Factor Approaches

Abstract

Improving the quality of predictions from sample data is an important aspect of a data collection and analysis program. Regional and environmental agencies use different statistical modeling techniques for prediction. The advent of neural network technology has introduced a new set of prediction models. It is important to make sure that the modeling technique used provides the best possible accuracy. This study compared the existing approaches to the estimations of average and peak hourly traffic volumes with the multiple regression analysis and the neural network approach. All the approaches were compared using different classification schemes as well as different durations of traffic counts. The multiple regression analysis and the neural network approaches consistently performed better than the conventional approach. Apart from suggesting good modeling tools for estimating average and peak hour traffic volumes, the results also provide useful insight into the durations of short-term traffic counts and the classification schemes for highway sites.