Optimized XGBoost-Based Framework for Robust Prediction of the Compressive Strength of Recycled Aggregate Concrete Incorporating Silica Fume, Slag, and Fly Ash

Yassir M. Abbas^1,*, Ammar Babiker², Fouad Ismail Ismail³
1 Department of Civil Engineering, College of Engineering, King Saud University, P.O. Box 800, Riyadh, 11421, Saudi Arabia
2 School of Civil Engineering, College of Engineering, Sudan University of Science and Technology, Eastern Daim, Khartoum, P.O. Box 72, Sudan
3 Department of Civil and Environmental Engineering, Florida International University, Miami, FL 33174, USA
* Corresponding Author: Yassir M. Abbas. Email: yabbas@ksu.edu.sa

Computer Modeling in Engineering & Sciences https://doi.org/10.32604/cmes.2025.074069

Received 01 October 2025; Accepted 12 November 2025; Published online 01 December 2025

Download PDF

Abstract

Accurately predicting the compressive strength of recycled aggregate concrete (RAC) incorporating supplementary cementitious materials (SCMs) remains a critical challenge due to the heterogeneous nature of recycled aggregates (RA) and the complex interactions among multiple binder constituents. This study advances the field by developing the most extensive and rigorously preprocessed database to date, which comprises 1243 RAC mixtures containing silica fume, fly ash, and ground-granulated blast-furnace slag. A hybrid, domain-informed machine-learning framework was then proposed, coupling optimized Extreme Gradient Boosting (XGBoost) with civil engineering expertise to capture the complex chemical and microstructural mechanisms that govern RAC performance. Systematic grid-search optimization (n_estimators = 50, learning_rate = 0.2, max_depth = 7) produced superior predictive accuracy (training R2 = 0.9923, testing R2 = 0.937; MAE = 2.378 MPa; RMSE = 3.591 MPa), which outperformed Extra Trees, Light Gradient Boosting, and traditional regressors. Beyond prediction, model interpretability was achieved using Shapley additive explanations and partial dependence analyses, which revealed curing age as the dominant strength driver, while water-to-binder ratio and recycled aggregate water absorption exhibited strong negative influences. Three-dimensional interaction plots further demonstrated how optimal superplasticizer dosages reduce the strength loss associated with high recycled aggregate content. In summary, this work provides a novel, explainable, and data-driven framework that achieves high predictive accuracy with mechanistic transparency and offers a powerful, interpretable tool for the design and optimization of sustainable RAC mixtures.