More eco-friendly and green concretes are needed to lower the climatic and environmental impacts of the growing demand for concrete. Despite the growing interest in using natural zeolite (NZ) in cement-based materials as an eco-friendly alternative for ordinary Portland cement (OPC), there is limited knowledge regarding the post-fire mechanical properties of natural zeolitic concrete (NZC). Hence, the development of a global model is desired to better explore the strength behavior of NZC after exposure to elevated temperatures. This study focused on the post-fire behavior of low-to-high-strength NZC specimens using a novel evolutionary method. Therefore, a widespread experimental program was designed to perfectly investigate the post-fire compressive strength of NZC. The experimental results were then used to develop an evolutionary-based machine learning (ML) model. To do so, the multivariate adaptive regression splines (MARS) method was hybridized with a new metaheuristic algorithm, the Horse herd Optimization Algorithm (HOA). Besides, the results of five different ML models, namely standalone MARS, M5P model tree (M5P), extreme learning machine (ELM), Gaussian process regression (GPR), and gene expression programming (GEP) were employed for comparison with the performance of metaheuristic-based MARS (Meta-MARS) model. Further external validation was conducted to show the superior performance of the Meta-MARS model. Also, a parametric study was conducted to demonstrate the robustness of the developed model. Moreover, an uncertainty analysis inspired by the Monte Carlo simulation (MCS) method was applied to the prediction results. Results of the modeling provided new insight into the post-fire behavior of NZC and agreed well with the experimental results. The satisfactory post-fire performance of NZC is promising and shows the climatic, environmental, and economic advantages of utilizing natural pozzolans.
