Application of Neural Networks to Estimate AADT on Low-Volume Roads

Abstract

Artificial neural networks are applied as a means of estimating the average annual daily traffic (AADT) volume from short-period traffic counts. Fifty-five automatic traffic recorder sites located on low-volume rural roads in Alberta, Canada are studied. The neural network models used in this study are based on a multilayered, feedforward, and back-propagation design for supervised learning. The AADT estimation errors resulting from various durations and frequencies of counts are analyzed by computing average and percentile errors. The results of this study indicate a clear preference for two 48-h counts as compared to other frequencies (one or three) or durations (24- or 72-h) of counts. In fact, the 95th percentile error values of about 25% for the neural network models compare favorably with the values reported in the literature for low-volume roads using the traditional factor approach. A number of advantages of the neural network approach over the traditional factor approach of AADT estimation are also included in the paper.