A comparative study of prediction models for alkali-activated materials to promote quick and economical adaptability in the building sector

Abstract

Alkali-activated materials (AAMs) have recently gained attention as potentially useful alternative binders that can reduce carbon dioxide emissions initiated by the production of Portland cement (PC) and decarbonize concrete production. Plenty of research has been conducted to explore the numerous characteristics of AAMs. However, when utilizing typical empirical and statistical methods, it is still difficult to accurately estimate the mechanical features of AAMs due to inaccuracy and uncertainty. This research presents machine learning (ML) models intending to estimate AAM's compressive strength (CS) based on their mixture's proportions. Literature records weremined to compile a comprehensive dataset with 676 mixture designs. Three ML models were employed to assess AAM's CS prediction capabilities. These models included an artificial neural network (ANN), a K-nearest neighbor (KNN), and a decision tree (DT). In addition, RreliefF and interaction analysis were conducted to examine the significance and interaction of various factors used in the database. Statistical assessments of results showed that ANN's prediction accuracy was higher than that of KNN and DT. The correlation coefficient (R2) values of 0.91 for ANN, 0.79 for KNN, and 0.78 for DT confirm the applicability of ANN for determining the CS of AAMs. RreliefF analysis showed that the most important variables were the amounts of sodium hydroxide (NaOH) and sodium silicate (Na2SiO3), followed by the fine aggregate, water/binder ratio, specimen age, and slag. The remaining variables moderately affected the results (i.e., CS of AAM). ML approaches can evaluate AAM CS across different input parameter values to optimize resource allocation and streamline future testing. The RreliefF analysis and interaction study indicated how input components affect AAM CS. These findings may help researchers and industry professionals determine the best raw material quantity, composition, and other factors for AAM development.