AI-driven optimization of slag-based geopolymer composites reinforced with basalt fiber using hybrid taguchi-GRA-ANN approach

Abstract This study presents an integrated experimental–predictive framework for the design and optimization of basalt fiber-reinforced slag-based geopolymer composites using a hybrid Taguchi–Grey Relational Analysis (GRA) and Artificial Neural Network (ANN) approach. The core scientific advancement lies in the explicit integration of experimental multi-response optimization with data-driven predictive modeling within a single coherent workflow. Taguchi–GRA is first employed to experimentally optimize compressive, flexural, split tensile, impact resistance, and water absorption properties simultaneously, capturing the inherent trade-offs among multiple performance criteria. The geopolymer matrix was synthesized using Ground Granulated Blast Furnace Slag (GGBFS) activated with an alkaline solution comprising NaOH and Na 2 SiO 3 . The experimentally obtained multi-response optimum is then used as a physically grounded reference to validate an independently trained ANN model, which captures nonlinear relationships between processing parameters and composite performance and enables reliable prediction beyond the tested experimental conditions. The ANN model, trained on an augmented dataset of 450 statistically validated samples, achieved a cross-validated prediction accuracy of 99.12% with dropout regularization to prevent overfitting. The optimal composite parameters were identified as 1.5% basalt fiber content, an alkali activator ratio of 2.0, and a curing temperature of 90 °C. Additionally, lifecycle and scalability considerations indicate potential reductions in CO 2 emissions and improved material recyclability, highlighting the practical relevance of the proposed framework for sustainable construction applications.