Utilizing clustering techniques in estimating traffic data input for pavement design

This paper presents an objective approach for establishing similarities in vehicle classification and axle load distributions between traffic data collection sites. It is based on clustering techniques that identify in succession groups of sites of decreasing similarity on the basis of the attributes specified (e.g., either the percentage of vehicles by class or the percentage of axles by load interval, respectively). This method is implemented in identifying clusters of sites with similar vehicle class distribution and axle load distributions, respectively. Extended coverage weigh-in-motion data (i.e., more than 299 days/year) from the long-term pavement performance database was used for this purpose. These data included 178 sites distributed through seven states. The paper explains the clustering methodology for one of these states and presents the clustering results for all seven states. This methodology allows estimation of traffic input to the new mechanistic-empirical pavement design guide from limited site-specific traffic data.