Data-driven Homogeneous Pavement Groups—Soft Versus Hard Clustering.

Rapid pavement deterioration is a major problem in areas with harsh weather exposure and/or high truck traffic loading. While several studies in recent years have focused on pavement deterioration and management systems, there is not, to date, a robust method explaining how large amounts of pavement data can be processed to identify homogeneous groups for the calibration of pavement performance models to support prioritization of maintenance, rehabilitation, or design of new pavements. This study employs machine learning to develop an approach, capable of partitioning pavement data with a close response to causal factors including traffic and weather conditions; and considering pavement performance through factors including international roughness index and deflections. Two classes of clustering, i.e. soft and hard partitioning, through K-means and Self Organizing Maps—SOM respectively, were tested to understand the correlation between daily factors and pavements deterioration. The goodness of clustering was tested using extrinsic and intrinsic evaluation methods. It was concluded from the results that SOM clustering provided reliable results, as it employs a soft clustering method, where data points can belong to more than one cluster at the same time. The results allowed better visualization of data points, identifying hidden relationships among clusters, and exploring correlations between external effects and homogeneous families. Moreover, it became apparent that unlike the common Markovian methods, including previous years' condition in the analysis of clusters has marginal to no effects on the detection of homogeneous grouping. The methodology proposed in this paper will produce more reliable pavement homogenous groups as input for performance models and decision-making for the maintenance and rehabilitation of pavements. Moreover, future researchers can use the results of this study to prepare a standardized design for pavement groups. [ABSTRACT FROM AUTHOR]