Elastic properties of rubber-modified asphaltic concrete under various stress states: characterization through explainable artificial intelligence (XAI) models

This study investigates the elastic properties of rubber-modified asphaltic concrete (RMAC) using explainable artificial intelligence (XAI) models. Laboratory cyclic loading tests were conducted on crumb rubber (CR)-modified asphaltic concrete (CRMAC, CR content 5–9%) and natural rubber latex (NRL)-modified asphaltic concrete (NRMAC, NRL content 3–7%) specimens, prepared using modified binders at 180 °C and 150 °C, respectively. The equivalent elastic Young’s modulus (Eeq) and equivalent Poisson’s ratio (νeq) were measured. An XAI framework utilizing a multilayer perceptron (MLP) neural network, interpreted through SHapley Additive exPlanations (SHAP), was adopted to predict and interpret these elastic properties. The optimized MLP model, identified through an ablation study, outperformed traditional machine learning algorithms in predicting Eeq and νeq. SHAP analysis revealed that stress level was the primary factor influencing both elastic properties, followed by rubber content and type. An optimal rubber content of 5−7% was identified for maximizing Eeq, with CRMAC mixtures generally displaying superior Eeq and νeq values compared to NRMAC mixtures. Non-linear relationships were observed, with Eeq increasing and νeq decreasing as stress levels rose. The proposed XAI approach introduces a novel, interpretable methodology for characterizing complex RMAC mixtures, offering valuable insights for optimizing mixture designs to enhance pavement durability and sustainability.