Modeling pavement condition index using Cascade Architecture: classical and neural network methods

Abstract

Currently, Palestinian municipalities define pavement conditions using detailed visual inspection. This procedure is expensive and time consuming. The new proposed model contributes to reducing both time and cost incurred by identifying a new model that considers machine-learning techniques rather than the conventional ASTM procedure. The authors studied the most common six pavement defects. These defects are alligator cracks, patching, longitudinal, and transverse cracks, shoving, and potholes. An optimized hybrid model is developed to determine the Pavement Condition Index (PCI) based on the FHWA Long-Term Pavement Performance (LTPP) database. The model follows a cascade architecture with three classical machine learning models followed by a neural network model. Using the neural network model as a second stage in the chain allows the model to focus its power on the nonlinear curve estimation to further reduce the errors. Finally, out-of-sample performance analysis as well as a cross-validation analysis is carried out to check the robustness of the model. The results show that the model estimates PCI with high degree of accuracy