Correlation-Based Clustering of Traffic Data for the Mechanistic–Empirical Pavement Design Guide

Development of traffic data clusters is crucial for use of the Mechanistic–Empirical Pavement Design Guide (MEPDG) when site-specific traffic data are not available and statewide data are too general. However, a preferred approach to traffic data clustering is not specified in the MEPDG. In current clustering practice, subjective decisions are made about issues such as determination of the number of clusters. This paper presents a new clustering combination method, correlation-based clustering, that considers the effects of traffic inputs on pavement design thicknesses, so that determination of the number of clusters is made objectively. For each traffic input required in the MEPDG, the similarity between two sites is evaluated with Pearson's correlation coefficient. Then, this approach evaluates the sensitivity of pavement design thickness to each traffic input to quantify locations to cut the hierarchical clustering trees, which objectively determines the number of clusters. The MEPDG requires many traffic inputs, including vehicle class distributions, four types of axle load spectra (per vehicle class), monthly and hourly distribution factors, and distributions of axle groups per vehicle. This clustering approach is performed for each traffic input so that a unique set of clusters can be developed for each traffic input. The method has been implemented for 22 direction-specific weigh-in-motion stations in Alabama to identify clusters of sites with similar estimated pavement performance for each traffic input of the MEPDG. This paper illustrates the clustering process for one traffic input (single-axle distribution) and presents clustering results for vehicle class distribution.