Abstract Manager

In recent years, Machine Learning has became a useful tool for several Civil Engineering applications, including Pavement Engineering. This study presents a two-part machine learning framework for analyzing and predicting pavement performance. The Long-Term Pavement Performance (LTPP) database was used for this study. For the first part, Random Forest regression was used for multi-target prediction of pavement indicators including International Roughness Index (IRI), rutting depth, fatigue cracking, and bleeding. The model integrates environmental parameters, crack severity metrics, and climate zone classifications. In the second part, IRI progression across climate zones was modeled by training Random Forest models on age-restricted (asphalt concrete pavement < 21 years) segment data. The analysis revealled distinct deterioration trends in wet-freeze, wet-nonfreeze, dry-nonfreeze, and maritime regions. Results showed strong predictive accuracy (R² > 0.9) and significant climate-based variation in roughness progression. The findings support the integration of climate zone and surface distress analytics into data-driven Pavement Management Systems (PMS).