Treffer: Machine Learning Prediction of Resilient Modulus for Base and Subbase Pavement Layers.

The resilient modulus (MR) is a critical parameter in pavement design, reflecting the elastic response of unbound materials under traffic loading. Accurate MR prediction reduces reliance on laboratory testing and enhances data-driven design. While machine learning has been applied to MR estimation, most studies focus exclusively on subgrade soils and overlook the distinct behaviors of base and subbase layers, as well as key material properties such as specific gravity, gradation, and hydraulic conductivity. This study addresses these gaps by developing separate machine learning models for base and subbase layers using 2019 cleaned records from the Long-Term Pavement Performance (LTPP) database and 17 input features. Feature selection based on a 1% importance threshold and hyperparameter tuning were applied. The Extra Trees model achieved the best performance for base layers (<italic>R</italic>2 = 0.86, MAE = 1 ksi), while K-Nearest Neighbors performed best for subbase layers (<italic>R</italic>2 = 0.88, MAE = 0.56 ksi), with cross-validation confirming model robustness. Prediction residuals showed strong agreement between observed and predicted MR values. A Python-based tool was also developed to estimate MR from user-defined material properties. These results demonstrate the potential of interpretable, layer-specific machine learning models to enhance MR prediction and improve mechanistic-empirical pavement design workflows. [ABSTRACT FROM AUTHOR]