Influence of chemical constituents of binder and activator in predicting compressive strength of fly ash-based geopolymer concrete using firefly-optimized hybrid ensemble machine learning model

This study focuses on investigating the impact of chemical constituents in binders and activators on the prediction of strength, particularly proposing a novel hybrid ensemble model that synergizes the strengths of three base models optimized using the Firefly Algorithm. The primary objective is to enhance the accuracy of predicting the compressive strength (CS) of fly ash-based (FA) geopolymer concrete (GPC). The dataset employed encompasses comprehensive material and chemical information, facilitating a predictive approach linking factors to strength. Rigorous preprocessing techniques are employed to eliminate outliers and ensure data integrity. Multivariate analyses are executed to visually represent the dataset's structure. The development of the hybrid ensemble model is realized through a stacking strategy, integrating the predictive capabilities of individual models. Thorough evaluation using diverse statistical metrics validates the superiority of the hybrid model compared to standalone base models, underscoring its enhanced precision in CS prediction for FA-based GPC. Sobol and FAST global sensitivity analysis was also employed to find the influence of input parameters on strength. Extra water content and curing temperature with Sobol indices of 67.2% and 42.5%, and FAST indices of 65.2% and 45% were found to be the most sensitive parameters of the studied database. Among the chemical constituents of binders and activators, the SiO2 content and CaO content of FA exhibited greater sensitivity, impacting the CS. On the other hand, the Na2O, Al2O3, and Fe2O3 content of fly ash and the SiO2 and Na2O percentage of sodium silicate were found to have a relatively lower impact.