Heavy metal contamination poses substantial environmental challenges, particularly as industrial operations continue to discharge toxic pollutants, threatening water quality and ecosystem health. A promising response to this crisis is Bentonite Plastic Concrete (BPC), which incorporates bentonite clay to absorb heavy metals effectively. Recent advances by researchers significantly improve how we can predict the properties related to BPC’s effectiveness, introducing hybrid ensemble learning models tuned by Forensic-Based Investigation Optimization (FBIO) to determine its workability and mechanical characteristics.
This approach merges various predictive models—including Random Forest (RF), Adaptive Boosting (ADB), Extreme Gradient Boosting (XGB), and Gradient Boosting Regression Trees (GBRT)—to achieve high predictive accuracy for properties like slump (S), tensile strength (TS), and elastic modulus (E). The study revealed specific input parameters integral to the predictions, including gravel, bentonite, silty clay, curing time, sand, cement, and water.
For example, the GBRT-FBIO model achieved astounding accuracy measures with R² values of 97.2% for elastic modulus predictions, underlining its reliability as engineers explore sustainable construction materials. Similarly, the XGB-FBIO model excelled at determining tensile strength and slump, reaching R² values of 97.7% and 96.6% respectively, which offers practitioners clear advantages.
Through Shapley Additive Explanation (SHAP) analysis, the research identified water as the most influential factor impacting slump predictions, contributing positively (+0.11) to model outcomes, followed by curing time, which also showed significant impacts on both tensile strength and elastic modulus predictions. These insights enable engineers to make informed decisions when designing concrete mixes, especially under varying environmental conditions.
Empowering engineers with real-time applications for these hybrid models, the authors created a user-friendly online tool, allowing practitioners to implement BPC predictions directly on-site—an advantage over traditional, often cumbersome testing methodologies. This innovative interface enhances efficiency and practicality, making it easier to determine the potential performance of various concrete mixtures.
The study’s conclusion advocates for these ensemble learning models as transformational tools within the field of civil engineering, particularly as they relate to sustainability practices surrounding construction materials. By incorporating advanced computational methods such as FBIO for hyperparameter optimization, the models not only improve prediction accuracy but also broaden the scope of applicability to real-world scenarios.
Despite its accomplishments, the research recognizes limitations, primarily stemming from the relatively small dataset sizes used for model training, which could inhibit broader generalizability. Future work aims to expand upon these models by integrating more diverse datasets and enhancing interpretability, ensuring the models remain relevant for complex construction needs.
Overall, this work marks notable progress toward utilizing technology to predict material performance reliably, enabling engineers to optimize concrete designs, ensuring structures are both effective and resilient against environmental adversities.