Development of a machine learning algorithm for predicting compressive strength of alkali-activated recycled concrete

Environmental challenges such as resource depletion, construction waste, carbon emissions, and the need for sustainable development have created an urgent demand for eco-friendly alternatives to traditional concrete. Alkali-activated recycled concrete (ARC) presents a promising solution by utilizing supplementary cementitious materials and construction and demolition waste as recycled aggregates; however, ARC lacks an accurate mix design methodology due to limited research. This study employs state-of-the-art machine learning (ML) models to predict the compressive strength (CS) of ARC. A dataset of 665 experimental results, with CS values ranging from 6.91 to 76.28 MPa, was compiled and preprocessed using variance inflation factor (VIF) analysis. Six ML models—BPNN, SVR, RF, GDBT, XGB, and LGB—were developed and optimized. XGB achieved the highest accuracy (R² = 0.995, MAE = 0.159 MPa, RMSE = 1.126 MPa, MSLE = 0.0004). SHAP analysis identified molarity and curing age as dominant parameters. A GUI validated using four ARC mixes showed prediction errors below 5%, confirming model reliability.