Assessing the impact of acid and sulfate exposure on alkali‐activated materials using metaheuristic machine learning

Abstract Climate change mitigation demands urgent shifts toward sustainable construction materials, with alkali‐activated materials (AAMs) emerging as a low‐carbon alternative to traditional cement. However, their durability in aggressive environments—critical for infrastructure resilience—remains understudied. This study develops four machine learning models (gene expression programming [GEP], K ‐nearest neighbors [KNN], support vector machine, and stochastic gradient descent [SGD]) to predict the strength loss (SL) in AAMs exposed to sulfate and acid attacks, using mix constituents and exposure duration as predictors. The SGD model achieved exceptional accuracy under acid attack ( R = 0.9, R 2 = 0.88), while GEP and KNN excelled in sulfate conditions ( R 2 = 0.9, R > 0.93). Shapley's analysis revealed that acid resistance hinges on minimizing the sand‐to‐binder ratio and exposure time, whereas sulfate durability depends on optimizing fly ash (inverse relationship with SL) and hydrated lime content (direct relationship). By linking these insights to environmental stressors intensified by climate change—such as acid rain and sulfate‐rich soils—the study provides a predictive framework to tailor AAM mixes for specific hazards. Engineers can leverage these models to design resilient infrastructure (e.g., marine structures and wastewater systems), reducing material degradation, maintenance costs, and lifecycle carbon footprints. This work bridges the gap between sustainable material innovation and climate‐resilient construction, advancing the adoption of AAMs in meeting global decarbonization goals.