Mechanical and durability performance prediction of geopolymer incorporating ferrosilicon slag and aluminum powder using machine learning techniques.

The present study explores the mechanical, durability, and environmental performance of geopolymer concrete (GPC) synthesized using ferrosilicon slag (FS) and aluminum powder (AP) as sustainable binder constituents. Eight mixes were prepared by varying FS: AP ratios (100:0, 95:5, 90:10, 85:15) and activating them with 6 M and 9 M sodium hydroxide at a constant activator-to-binder ratio. Experimental tests included compressive, flexural, and split tensile strengths, water sorptivity, and rapid chloride penetration (RCPT). Results showed that increasing AP content improved workability up to an optimum of 10%, after which excessive porosity reduced strength. The 90% FS–10% AP mix at 9 M NaOH (M7) achieved the highest performance, yielding 53.2% higher compressive strength, 23.8% higher flexural strength, and 24.0% higher split tensile strength than the corresponding 6 M mix. Durability also improved significantly, with sorptivity reduced by 25.0% and RCPT charge by 52.9% at higher molarity. Machine learning (ML) models artificial neural networks (ANN), random forest (RF), and support vector regression (SVR) were trained to predict compressive strength, with RF achieving the best accuracy (R2 = 0.98). A cradle-to-gate carbon footprint analysis demonstrated that AP-free mixes had the lowest embodied CO2, whereas the optimal M7 mix provided the best balance between performance and sustainability. Overall, the study highlights the synergistic potential of FS and AP in geopolymer concretes and provides an integrated experimental–ML–carbon framework for designing high-performance, low-carbon GPC mixtures.