Advanced predictive techniques for estimating compressive strength in recycled aggregate concrete: exploring interaction, quadratic models, ANN, and M5P across strength classes

Concrete is one of the most versatile composite materials used in construction on a large scale. A by-product of concrete production is recycled aggregate. Recycling aggregate helps reduce the consumption of natural resources and minimizes the amount of waste sent to landfills, contributing to environmental sustainability. Recycled coarse aggregate (RCA) concrete has been used as a partial or full replacement for gravel, aiming to maintain or improve key mechanical properties, such as compressive strength (CS). Compressive strength becoming increasingly critical and is prioritized early in the construction design process to ensure structural integrity. A mathematical model is essential for accurately predicting the compressive strength of recycled aggregate concrete. This study evaluates predictive models for estimating the compressive strength of recycled aggregate concrete (RAC) based on 708 experimental results from the literature. The considered models are the interaction model, full quadratic model (FQ), artificial neural network (ANN), and M5P-tree. The modeling process was focused on the key factors affecting the compressive strength of concrete when RCA was used as a substitute. These variables include the water-to-cement ratio from 0.25 to 0.97, cement content from 185 to 864 kg/m³, gravel content from 0 to 1436 kg/m³, gravel size from 4.75 to 37.5 mm, sand content from 363 to 1105 kg/m³, Superplasticizer from 0 to 4.28%, curing time from 1 to 365 days, and recycled coarse aggregates from 0 to 1393.31 kg/m³. According to statistical analysis, the ANN model achieved the highest predictive accuracy with an R² of 0.965 and a low root mean square error (RMSE) of 4.32 MPa, significantly outperforming other models like the interaction model and M5P-tree.