Development of machine-learning performance prediction models for asphalt mixtures

The primary objective of this study is to investigate the effect of in-place air voids (AV), asphalt content (AC), bulk-specific gravity (BSG), and maximum specific gravity of asphalt (G_mm) on the fatigue cracking of asphalt mixtures using the Long-Term Pavement Performance (LTPP) database. This study includes 13 sections from different locations covering different mix designs, pavement age between 20-30 years, and two climate zones across the United States. All data were derived from LTPP database. A multiple linear regression, random forest (RF) and support vector machine (SVM) methods between the selected explanatory properties and the fatigue cracking were used for the investigation. A multiple significant linear regression model was developed, and it confirmed the significant relationships between the AC, AV, G_mm, percent of aggregate Pass.No.200, BSG and the fatigue cracking. RF and SVM methods were established using the same data. They validated significant properties and the accuracy of the model.