The mechanical properties of recycled brick aggregate concrete at elevated temperatures: Ensemble machine learning models and experimental validation

This study evaluated the mechanical properties (MP) of recycled brick aggregate concrete (RBC) at elevated temperatures using both computational prediction models and experimental validation‏. The employed prediction models include Random Forest (RF), Gradient Boosting (GB), XGBoost (XGB), CatBoost (CB), and Stacking (ST). A comprehensive database was compiled to train and test the models. These data points included the mixed design parameters of RBC and exposure temperatures as inputs. The outputs consisted of residual compressive strength (RSC), flexural strength (FS), modulus of elasticity (ME), and Ultrasonic Pulse Velocity (UPV). Random sampling and 5-fold cross-validation (CV) techniques were applied to select the best ML model for predicting RCS. The ST model exhibited the highest predictive performance for RCS, FS, ME, and UPV prediction on the test data, achieving coefficients of determination (R²) of 0.931, 0.936, 0.944, and 0.972, respectively. SHAP sensitivity analysis was conducted to evaluate the impact of each input variable on the predicted MP, revealing that temperature was the most influential input variable across all outputs. To enhance the evaluation of machine learning models, RBC samples with specific mix designs were prepared, and RCS was measured at elevated temperatures. The results showed that the relative error in all cases remains below 17.54%. Finally, a web-based application was ultimately developed that allows users to predict the MP of RBC using https://mp-of-rbc-in-high-temp.streamlit.app/ . • ML models predicted the mechanical properties of recycled brick aggregate concrete. • The ST model achieved the best performance on the test data across all outputs. • SHAP analysis identified temperature as the most influential factor. • A application was designed and developed based on the findings of this study. • Experimental validation confirmed the high accuracy of the application.