Effect of Data Imbalance on the Performance of Pavement Deterioration Models

Abstract

Pavement performance models that can model consistently well across different pavement types, structures, traffic, and environmental conditions as well as throughout pavements’ lives are essential for the efficient management of transportation infrastructures. However, network pavement performance data are inherently imbalanced, and most pavement performance models are built on algorithms that assume balanced data distribution and equal error cost. As a result, the performance of most pavement performance models may vary for different pavement groups. Considering this, the objective of this study was to evaluate the data imbalance of the network pavement performance data and its effect on the performance of pavement deterioration models. Specifically, the analysis result for the longitudinal cracking model is presented in this paper. The data imbalance of the network pavement performance data mainly arises from two sources: the intrinsic pavement characteristics and pavement life expectancies. The intrinsic data imbalance arises either from the design or the management process, whereas the data imbalance introduced by the pavement life expectancies is primarily the result of the pavement management process. The distribution of the data used in this study exhibits substantial skewness owing to the data imbalance caused by the intrinsic pavement characteristics and life expectancies. The existence of the data imbalance leads to the biased performance of the deterioration models toward pavement classes with more data.