Parametric study of pavement deterioration using machine learning algorithms

The long-term pavement performance (LTPP) database is a valuable resource for studying the performance of different pavement structures under various environmental and traffic conditions. The database could be employed in developing performance prediction models and help to prioritize the maintenance strategies. Soft computing techniques can be utilized to evaluate the performance of quality control programs (QCP) during the material production and construction processes. These data techniques can provide relationships between QCP parameters and the corresponding long-term performance indicators such as permanent deformation, roughness, and cracking. This paper employs the LTPP database, comprised of the measured quality control parameters, such as voids in mineral aggregates (VMA), air voids of the mixture (VA), in-place density of asphalt concrete, and the age of pavement structure, as well as deterioration indices. A hybrid machine learning (ML) method that combines random forest (RF) and artificial neural network (ANN) was developed for the prediction of alligator deterioration index (ADI). The model was then used to conduct a parametric study using a wide range of independent variables to investigate how they affect the ADI. The results showed that the hybrid ML technique is capable of predicting pavement deterioration rigorously.