Predicting compressive strength of recycled aggregate concrete using AI with experimental validation

Abstract

Growing concerns about environmental impact and rising construction costs have encouraged the use of recycled aggregates in concrete. This study offers an experimental and AI-driven assessment of the compressive strength of recycled aggregate concrete. Concrete was prepared by varying recycled fine and coarse aggregates, water-cement ratios, plasticizer content, and parent concrete strength. The measured 28-day compressive strength ranged from 31.8 to 45.2 MPa, and a steady reduction was observed as recycled aggregate levels and water absorption increased. The results highlight how aggregate quality, abrasion resistance, and replacement percentage influence mechanical performance. To extend the laboratory findings, nine machine learning models were developed to estimate compressive strength based on all mix parameters. Support Vector Regression delivered the strongest performance with an R2 of 0.998, followed by Random Forest (0.996). K-Nearest Neighbors and XGBoost achieved R2 values of 0.947 and 0.942, while Gradient Boosting, Linear Regression, Lasso, Elastic Net, and Ridge obtained R2 scores of 0.904, 0.893, 0.879, 0.846, and 0.842. Nonlinear algorithms clearly performed better than linear models. The results show that AI tools can reliably predict compressive strength, reducing the need for extensive testing and helping streamline mix design for recycled aggregate concrete.