Compressive strength evaluation and predictive modeling of mortar incorporating zirconia and rice husk ash using ensemble machine learning

Abstract

In the context of sustainable construction, this study quantifies and predicts the compressive strength of cement mortar incorporating zirconia (ZrO₂) as a partial cement replacement (1–5% by mass) and rice husk ash (RHA) as a partial sand replacement (10–50% by mass). Mortars were proportioned at 1:3 (binder:sand) with a constant w/c = 0.50; 70.6 mm cubes were cast and tested at 3, 7, 14, 28, and 56 days following IS 516:2018. The best-performing mix within the investigated range—3% ZrO₂ and 10% RHA—exhibited a 33.8% increase in 28-day compressive strength relative to the control, attributed to zirconia’s micro-filling/nucleation effects and the pozzolanic reactivity of RHA. To predict strength from mix proportions and curing age, four machine-learning models (CatBoost, XGBoost, Random Forest, LightGBM) were trained using tenfold cross-validation on 130 experimental observations and evaluated on a held-out test set. CatBoost and XGBoost achieved Test R2 = 0.998 (RMSE ≈ 0.30 MPa), Random Forest Test R2 = 0.990 (RMSE ≈ 0.75 MPa), and LightGBM Test R2 = 0.975 (RMSE ≈ 1.15 MPa). Feature-importance analyses consistently identified curing age as the most influential variable in strength development. The results demonstrate a reproducible laboratory–data workflow for evaluating and predicting the strength of ZrO₂–RHA mortars without claiming mix-design optimization beyond the tested ranges.