Abstract
This study investigates the mechanical behavior and predictive modeling of cementitious mixtures partially substituted with Reclaimed Asphalt Pavement (RAP), aiming to assess the feasibility of RAP in sustainable concrete applications and the utility of machine learning (ML) in forecasting performance outcomes. Experimental trials were conducted using concrete mixes with varying RAP contents (0%, 25%, and 50%) under controlled water-to-cement ratios and curing durations. Key performance metrics-including compressive strength, tensile strength, density, and water absorption-were measured to evaluate the influence of RAP on structural integrity. To predict compressive strength based on mix parameters, two supervised ML algorithms-Random Forest Regression (RFR) and Support Vector Regression (SVR)-were applied. Despite the modest dataset, SVR demonstrated superior predictive accuracy, achieving lower Root Mean Squared Error (RMSE = 2.21 MPa) and Mean Absolute Error (MAE = 1.99 MPa) compared to RFR (RMSE = 2.88 MPa; MAE = 2.74 MPa). Notably, compressive strength declined significantly as RAP content increased, accompanied by higher water absorption and reduced density. These findings highlight the trade-off between mechanical performance and sustainability in RAP-modified concretes while validating the potential of artificial intelligence for materials design. ML models effectively captured nonlinear behaviors, even with limited data, enabling scalable optimization of sustainable concrete formulations.