An Unsupervised Cluster Method for Pavement Grouping Based on Multidimensional Performance Data

Abstract

Pavement condition data usually consist of multiple performance indicators, which poses difficulties in maintenance and fund allocation decision making. In conventional practice, empirical-based scoring methods have been used to assess multidimensional pavement condition attributes. However, if performance data on a roadway network are available, data-driven approaches can be applied for such multiattribute decision-making problems. In this study, an unsupervised cluster method called normalized cuts (NCut) is developed to group pavement sections into clusters with homogenous conditions. Geometric centers of the clusters are used to determine the performance ranking of each cluster. The proposed methodology is demonstrated with a case study in Louisiana. A total of 35 pavement sections with eight performance parameters are grouped into five clusters indicating conditions ranging from very good to very poor. It is validated with current practice that the methodology presented in this study is effective in supporting pavement prioritization decision making.