Machine learning-driven evaluation of mechanical and microstructural properties of agro-waste-derived geopolymer...

The growing demand for sustainable construction materials has accelerated research into eco-friendly alternatives to traditional Portland cement. This research explores the potential of geopolymer concrete formulated from agricultural-waste ashes as a sustainable replacement for conventional Portland cement. Banana peel ash (BPA) and sugarcane bagasse ash (SCBA) were employed as aluminosilicate precursors, and their combined effects were systematically examined through controlled variations in blend proportion, alkaline activator molarity, sodium silicate-to-sodium hydroxide (SS/SH) ratio, and aggregate-to-binder ratio. The influence of these parameters on fresh and hardened properties-including workability, compressive strength, and flexural strength-was rigorously evaluated. Within the defined experimental domain, an optimal formulation comprising 52.5% SCBA and 47.5% BPA activated with 10 M NaOH achieved compressive and flexural strengths of 33.17 MPa and 9.95 MPa, respectively, demonstrating structural-grade performance suitable for practical applications. Detailed microstructural investigations employing SEM-EDS, XRD, FTIR and TGA techniques confirmed that both ashes exhibit high silica content, significant pozzolanic behaviour, and that increased activator concentration enhanced the dissolution of aluminosilicate phases leading to a denser geopolymeric matrix with improved durability. To further strengthen the analytical framework and enable predictive mix optimization, artificial intelligence-based models-Gene Expression Programming (GEP) and Artificial Neural Networks (ANN)-were developed. Both models achieved excellent predictive performance (R² > 0.98) with respect to slump Flexural and compressive strength; however, the GEP model consistently exhibited superior accuracy, lower error indices and better alignment with measured results than the ANN. Performance was validated through statistical metrics including Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and coefficient of determination (R²), confirming the robustness of the machine-learning framework in capturing the complex, non-linear interactions among mix variables. The novelty of this study lies in demonstrating that low-value agricultural waste ashes can be engineered into reliable, structural-grade geopolymer binders, producing a high-performance BPA-SCBA concrete that repurposes agricultural residues and reduces the carbon footprint of cement production. Additionally, the integration of AI-based optimization provides a robust decision-support tool for mix design, enabling data-driven, sustainable construction practices.