Efficient compressive strength prediction of concrete incorporating recycled coarse aggregate using Newton’s boosted backpropagation neural network (NB-BPNN)

Abstract

This study advances the field of concrete compressive strength prediction by introducing an innovative approach incorporating recycled coarse aggregates and the Newton's Boosted Backpropagation Neural Network (NB-BPNN) model. Initial statistical validation of the dataset ensures its integrity for subsequent model training. The dataset analysis reveals variations in the usage of essential ingredients, with cement playing a pivotal role and showing a substantial range of 338.57 kg/m³ . Robust data preprocessing, including outlier handling, proved critical, substantially enhancing machine learning model performance. Post-outlier treatment, XGBoost and MLP models improved, while the decision tree model's performance saw a minor decrease. The NB-BPNN model's effectiveness in predicting concrete compressive strength was a cornerstone of this research. It demonstrated a remarkable average R² score of 0.95, explaining about 95% of the variance in compressive strength. The RMSE and MAE values of 3.205 MPa and 2.349 MPa, respectively, validated its accuracy and practical utility. Moreover, comparative evaluations across different compressive strength ranges emphasized the NB-BPNN model's superiority. While other models yielded varying results, the NB-BPNN consistently performed well, highlighting its efficiency and unique advantages in this domain. Feature importance analysis, through Partial Dependence Analysis (PDA) and SHapley Additive exPlanations (SHAP), identified key variables influencing the NB-BPNN model. Variables such as Cement, GGBS, Binder, Superplasticizer, and Water/Binder were found to significantly impact compressive strength, aligning with domain knowledge. This study underscores the NB-BPNN model's potential to revolutionize concrete compressive strength prediction and encourages further research in optimizing concrete mix design and construction practices.