Experimental and machine learning prediction of compressive strength of chemically activated RHA based RAC using SHAP and PDP analysis

Abstract

Abstract The increasing demand for sustainable construction materials necessitates the effective reuse of industrial and agricultural waste in high-performance concrete (HPC). However, challenges such as strength loss due to recycled concrete aggregates (RCA) and variable performance of supplementary cementitious materials hinder widespread adoption. This study addresses these challenges by investigating the synergistic effect of chemically activated rice husk ash (RHA), RCA (0–100%), and foundry sand on the compressive strength and durability of HPC. Six experimental groups were prepared: one with inactivated RHA and five with chemically activated RHA using 3.5% sodium sulfate (Na 2 SO 4 ), combined with RCA replacement levels of 0%, 40%, 60%, 80%, and 100%. All mixes included 20% FS as partial fine aggregate replacement and constant silica fume. Compressive strength was measured at 3, 7, 14, 28, 56, 90, and 120 days, while durability was evaluated through acid exposure tests over 4 months. To complement the experimental study, machine learning models including K-Nearest Neighbors, Random Forest, Artificial Neural Networks, and Extreme Gradient Boosting were applied to predict compressive strength. Among them, XGB outperformed others with an R 2 of 0.951, RMSE of 3.222 MPa, and MAE of 1.862 MPa. SHAP and Partial Dependence Plot (PDP) analyses revealed curing age, RCA, and Na 2 SO 4 content as key influencing factors. This study concludes that up to 40% RCA can be effectively used in HPC with activated RHA and FS without compromising long-term strength and acid resistance. The integration of interpretable ML models with detailed experimental validation provides a robust framework for sustainable concrete design.