Integrated experimental and machine learning study on sustainable concrete incorporating fly ash and eco-sand

The construction industry faces increasing pressure to reduce carbon dioxide emissions and minimise dependence on conventional cement and natural fine aggregates, motivating the development of sustainable concretes incorporating industrial by-products. In this study, the combined use of fly ash (FA) and eco-sand (ES) as partial replacements for cement (0–25%) and fine aggregate (0–20%) was evaluated across 26 concrete mix designs. Experimental assessments included mechanical, durability and microstructural characterisation through tests for workability, strength, water absorption, porosity and acid resistance, along with scanning electron microscopy/energy-dispersive spectroscopy analysis. The results showed that the mix with 15% FA and 10% ES achieved the highest compressive strength of 39.1 MPa, representing a 10.8% improvement over the control mix. The mix with 20% FA and 15% ES exhibited enhanced durability with reduced porosity (6.5%), water absorption (1.42%) and acid-induced mass loss. To extend the experimental investigation and enable predictive performance assessment, machine learning models (including support vector regression with radial basis function (SVR-RBF), bagging SVR and Gaussian process regression) were developed using a 70:30 training–testing split and leave-one-out cross-validation. SVR-RBF achieved the highest predictive accuracy (coefficient of determination of 0.9801 and root mean squared error of 0.349 MPa), although interpretation should consider the limited dataset size (n = 26). Shapley additive explanations analysis identified ES content and bulk density as dominant predictors, and a graphical user interface was developed for real-time prediction.