Mixture Optimization of Mechanical, Economical, and Environmental Objectives for a Novel Industrial Waste–Based Geopolymer: Combining Ensemble Learning with Metaheuristic Algorithms

A novel industrial waste–based geopolymer prepared with silica fume–based activator is more effective and cost-efficient than traditional geopolymers, but is not widely used due to insufficient research on optimized mix proportions and correlations between preparation parameters and compressive strength. An improved ensemble model, combining seven machine learning (ML) models with an isolated forest algorithm and elimination mechanisms, was developed to predict the compressive strength of silica fume–activated slag–fly ash geopolymers (SF-SFGs). The influence of preparation conditions and mix proportions was analyzed using variable importance (VI) and partial dependence plots (PDPs). Based on the improved ensemble model and a metaheuristic algorithm selected by performance comparison, the optimized mix proportions of SF-SFG considering compressive strength, cost, greenhouse gas emissions and energy consumption were obtained in three cases to guide its application. The results show that the improved ensemble model has superior accuracy and generalization compared with other individual models, with an R2 of 0.984, mean absolute error (MAE) of 2.351 MPa, and RMS error (RMSE) of 3.034 MPa, an increase of 2.4%–7.8%. The main variables influencing the compressive strength of SF-SFG are the activator dosage (30.72%), liquid–solid ratio (21.11%) and activator modulus (19.67%). The dung beetle optimizer (DBO) algorithm demonstrated strong efficacy in optimizing the mix proportions of SF-SFG. Compared with cement paste and traditional geopolymers, the manufacturing cost, greenhouse gas emissions, and energy consumption of SF-SFG with optimized mix proportions are reduced by approximately 21.3%–48.0%, 23.4%–86.7% and 33.2%–77.7%, respectively.