Machine Learning-Based Prediction of Mechanical Properties of Recycled Aggregate Concrete

Recycled Aggregate Concrete (RAC) has become an important material in sustainable construction due to its ability to utilize construction and demolition waste. However, its mechanical behavior is highly variable because recycled aggregates contain adhered mortar, impurities, and irregular surface textures. These inconsistencies make the prediction of compressive strength, splitting tensile strength, and flexural strength difficult using traditional empirical formulas. This paper presents a machine-learning-based framework for predicting the mechanical properties of RAC using XGBoost. A curated dataset consisting of experimental and literature-based RAC mix proportions was preprocessed, analyzed, and trained to construct three predictive models. The results show that XGBoost captures nonlinear relationships effectively and delivers higher accuracy than baseline methods such as SVM and KNN. The model demonstrates strong generalization and low prediction error, proving its suitability for mix-design optimization and rapid decision-making in civil engineering.