Roadway Seasonal Classification Using Neural Networks

Abstract

The parameter average annual daily traffic (AADT) is an important factor used in transportation planning, design, maintenance, and other transportation decision-making processes. The most commonly used approach for obtaining the AADT of a particular roadway facility is to obtain short-term counts (usually 48 h), apply the appropriate seasonal coefficients (also known as seasonal factors), and compute the AADT. In this study, a thorough discussion regarding AADT parameter, seasonal factor, annual traffic pattern, and road attributes related to roadway systems is presented. The existing approaches for determining traffic pattern—cluster analysis and regression analysis—are reviewed. Subsequently, an investigation of the potential applicability of the adaptive resonance theory 1 type of neural network associated with roadway classification and pattern recognition is given. In the comparative-analysis part of this paper, all three methods are applied to the same database, and the results of each method's performance are presented. Finally, after discussing the results, the conclusion is made that the presented adaptive resonance theory's performance for the given problem is superior when compared to the conventional methods. Some thoughts for further research in this area are also presented.