AI-Enhanced Data-Driven Approach to Model the Mechanical Behavior of Sustainable Geopolymer Concrete

Abstract

The growing environmental burden of ordinary portland cement has accelerated the search for sustainable alternatives. Geopolymer concrete (GPC), synthesized from industrial byproducts such as fly ash and slag, is a promising candidate. However, its compressive strength is difficult to predict because numerous mix and curing variables interact in complex, nonlinear ways. This study develops and validates an interpretable machine learning framework for forecasting GPC compressive strength. A curated data set of 350 samples was used, containing eight independent variables, i.e., fly ash, ground granulated blast-furnace slag, NaOH molarity, coarse aggregate, recycled aggregate, curing temperature, curing time, and age of testing, and one dependent variable, i.e., compressive strength. Six models (linear regression, decision tree, gradient boosting, support vector regression, artificial neural networks, and random forest) were trained and optimized. Their outputs were integrated into a stacking ensemble, i.e., the super learner (SL) model, with random forest as the metalearner. Performance was evaluated using R 2 , root mean square error (RMSE), and mean absolute percentage error (MAPE) under 10-fold cross-validation and a held-out test set. The SL achieved R 2 = 0.994 , RMSE = 0.90    MPa , and MAPE = 1.4 % , consistently outperforming all base learners and surpassing results reported in recent studies when benchmarked under comparable variable ranges. Model interpretability was ensured through Shapley Additive exPlanations and permutation sensitivity analyses, which identified curing time, NaOH concentration, test age, and curing temperature as dominant predictors. The novelty of this work lies in demonstrating that a stacking ensemble with transparent feature attribution can markedly improve accuracy and trustworthiness in predicting GPC strength, offering a rapid and nondestructive tool for sustainable mix design optimization.