Integrating obsidian and silica fume in geopolymer mortars: Strength prediction via meta-ensemble machine learning framework

Renowned for its durability and structural strength, concrete revives to lead global construction as the material of choice. However, the carbon-intensive nature of cement production demands the pursuit of greener, more sustainable alternatives. Geopolymer mortars derived from industrial by-products like obsidian (OB) and silica fume (SF) offer a sustainable alternative to conventional binders, but accurately assessing their behavior under diverse curing regimes remains a significant challenge. Furthermore, although there are many studies on machine learning (ML) methods and different types of geopolymer in the literature, there is no comprehensive study on predicting the compressive strength of geopolymers containing OB (90–100 %) and SF (0–10 %) using ML-based methods. This study therefore aims to address this gap by predicting the compressive strength of a dataset consisting of 150 data points created by varying the OB and SF ratios. The current research offers a robust ML framework for strength prediction of geopolymer mortars featuring OB and SF additives. Five popular ML techniques covering Gaussian Process Regression, Extremely Randomized Trees, Extreme Gradient Boosting, Bagging, and Decision Tree were tested both individually and in combination through a hybrid meta-model. The combined model delivered the best results, reaching an R 2 of 0.979, outperforming the standalone models, which scored between 0.87 and 0.963. The principal factors such as the proportions of OB and SF, curing temperature, and curing duration were examined using Feature Importance and Permutation Feature Importance analyses, with ANOVA confirming their relevance. K-fold cross-validation verified model's robustness, demonstrating ensemble ML methods substantially improve the precision and reliability of strength predictions for geopolymer mortars. These findings advance the design of sustainable construction materials while contributing to reduced carbon emissions in the building industry. • An innovative model was used to predict the CS of geopolymer mortars. • The Decision Tree performed best on its own, while the meta-model performed better. • The most important factor on the prediction of CS is the curing time. • ML algorithms are suitable for the prediction of features of geopolymers.