Modeling the Compressive Strength of Metakaolin-Based Self-Healing Geopolymer Concrete Using Machine Learning Models

Metakaolin-based self-healing geopolymer concrete treated with Bacillus bacteria represents a significant advancement in sustainable construction due to its eco-friendly properties, enhanced durability, and self-healing capabilities. It is a transformative material for sustainable construction. By reducing carbon emissions, utilizing waste, improving durability, and lowering lifecycle costs, it aligns with global goals for environmentally friendly and resilient infrastructure. Continued research and development will further unlock its potential, making it a cornerstone of the future of sustainable construction. In this research project, a study on modeling the compressive strength of environmentally friendly metakaolin-based self-healing geopolymer concrete treated with Bacillus bacteria (BB) has been conducted, analyzed, and reported. Machine learning methods such as the “Group Methods Data Handling Neural Network (GMDH-NN)”, “Generalized Support Vector Regression (GSVR), “K-Nearest Neighbors (KNN)”, “Tree Decision (Tree)”, “Random Forest (RF)” and “Extreme Gradient Boosting (XGBoost)” were applied to model the compressive strength of the self-healing concrete. The GMDH-NN model was created using GMDH Shell ۳.۰ software, while XGBoost, GSVR, KNN, Tree, and RF models were created using “Orange Data Mining” software version ۳.۳۶. The research method also included gathering relevant experimental and field data, categorizing it effectively, and performing initial analysis to identify trends and relationships. A global representative database was collected from literature for different mixing ratios of self-healing concrete corresponding to the compressive strength, with a total of ۱۴۷ records, which contained Fly Ash (FA), Silica Fume (SF), Metakaolin (MK), and Bacillus Bacteria (BB) considered as the input constituents. The collected records were divided into a training set (۷۵%) and a validation set (۲۵%) based on established requirements. At the end of the modeling exercise, the GMDH-NN produced the best model with an accuracy of ۰.۹۹, while the KNN and the GSVR followed closely with accuracies of ۰.۹۷۵ and ۰.۹۷, respectively. However, the RF and the Tree models also produced good accuracies of ۰.۹۶۵ and ۰.۹۵۵, respectively. Also, the GMDH-NN and the KNN again outperformed the other methods, producing an R² of ۱.۰۰ and ۰.۹۹, respectively, while the GSVR, RF, and Tree followed in this order with R² of ۰.۹۸, ۰.۹۷, and ۰.۹۶, respectively. The error indices, such as the overall error, RMSE, MSE, MAE, and SSE, also confirm this order of performance. The sensitivity analysis on the modeling of compressive strength of metakaolin-based self-healing geopolymer concrete treated with Bacillus bacteria produced a metakaolin (MK) impact of ۳۰%, a silica fume (SF) impact of ۲۹%, a fly ash (FA) impact of ۲۷%, and a Bacillus bacteria (BB) impact of ۱۴%. This highlights the dominant role of metakaolin (۳۰%), silica fume (۲۹%), and fly ash (۲۷%) in determining the compressive strength of metakaolin-based self-healing geopolymer concrete. Bacillus bacteria (۱۴%) have a smaller but meaningful impact, primarily contributing to self-healing and long-term durability. These insights can guide material selection, mix design, and process optimization to enhance both strength and durability. Doi: ۱۰.۲۸۹۹۱/CEJ-۲۰۲۵-۰۱۱-۰۴-۰۲۰ Full Text: PDF

کلیدواژه ها:

Self-healing Geopolymer Concrete ، Bacillus Bacteria ، Compressive Strength ، Metakaolin ، Machine Learning ، Green Concrete.