Multi-scale evaluation of waste glass as a fine aggregate replacement in asphalt mixtures: machine learning interpretation and experimental characterization of performance

Abstract With growing environmental and regulatory emphasis on sustainable construction, integrating recycled materials into asphalt mixtures has become an essential direction for advancing greener pavement technologies. This study aims to evaluate the feasibility and mechanical performance of asphalt concrete mixtures incorporating waste glass as a partial replacement of fine aggregates by weight at substitution levels of 0%, 10%, and 15%. Laboratory testing was conducted to assess stiffness, viscoelastic behavior, and resistance to permanent deformation through dynamic modulus and Flow Number measurements under varying temperatures and loading frequencies. To move beyond purely empirical evaluation, the research combines experimental characterization with interpretable machine learning to explore the complex interactions among mixture composition, temperature, and loading conditions. Eight machine learning algorithms were compared to identify the most influential parameters and to model nonlinear relationships between glass content, binder characteristics, and volumetric properties. Results showed that mixtures containing 10% waste glass exhibited up to a 15% increase in stiffness and improved rutting resistance, whereas higher replacement levels slightly reduced flexibility. Temperature and frequency were confirmed as the dominant factors governing modulus behavior. By linking experimental findings with data-driven interpretation, this work offers a transparent framework for understanding the multi-scale mechanisms governing waste glass–modified asphalt mixtures. The findings provide practical guidance for optimizing sustainable pavement mix designs, supporting informed decision-making in the broader shift toward circular construction practice.