Experimental assessment and data-driven hybrid machine learning quantification with parametric optimization of compressive strength of ceramic waste concrete

Abstract

Rising industrialization has increased ceramic waste generation and posed a significant environmental footprint due to disposal issues. This study aims to develop hybrid machine learning (ML) models for ceramic waste-based concrete (CWC) to predict its compressive strength (CS) and analyze the mix parameters that significantly impact its strength. For this purpose, 397 experimental data from published research were analyzed systematically, and six models—Decision Tree (DT), Bagging, Backpropagation Neural Network (BPNN) and their hybrid forms (BPNN-DT, BPNN-Bagging, and Bagging-DT) were constructed to predict the CS. CWC samples were prepared for experimental evaluation by replacing cement with 5 % SF and 5 %, 10 %, 15 %, and 20 % CW. A compressive strength test was conducted to assess the mechanical strength, and microstructural analysis was performed to highlight the interplay between CW and the concrete matrix. Furthermore, SHAP analysis and partial dependence plots (PDP) were used to show how each input feature affected the CS. According to the results, the hybrid Bagging-DT model outperformed other models and obtained training and testing R2 values of 0.9965 and 0.9623, with training RMSE, MAE, and MAPE of 1.252, 0.678, and 2.667 %. The feature importance analysis identified water, curing age, and superplasticizer as the most influential factors in predicting the CS of CWC. Microstructural analysis revealed that the incorporation of CW showed denser microstructure and greater C-S-H bonds, which enhanced compressive strength. The outcomes of this study will help the sustainable manufacturing of concrete by utilizing ceramic waste. At the same time, ML modeling will accelerate advancements by reducing labor-intensive and time-consuming experimental trials.

Keywords

• Ceramic waste
• Machine learning
• Compressive strength
• Parametric analysis
• Microstructural analysis