A recent study has explored the use of advanced machine learning (ML) techniques to evaluate the compressive strength of concrete made with industrial waste and reinforced with steel fibers.
Study: Evaluating the strength of industrial wastes-based concrete reinforced with steel fiber using advanced machine learning. Image Credit: Javidestock/Shutterstock.com
The researchers applied a range of ML models—including Kstar (a semi-supervised classifier), M5Rules (a rule-based model), ElasticNet, correlated Nystrom views (XNV), and decision tables (DT)—to predict compressive strength with high precision.
Background
Recycling industrial waste into construction materials offers a dual benefit: it reduces environmental impact while enhancing structural performance. Steel fibers, often discarded during manufacturing, can be repurposed to strengthen concrete, improving its tensile strength, crack resistance, and durability. This extends the lifespan of infrastructure and reduces the need for frequent repairs.
However, traditional testing methods for compressive strength are costly, time-intensive, and resource-heavy. That’s where machine learning comes in. By applying ML techniques, researchers can model and predict concrete performance more efficiently—particularly when working with unconventional or recycled materials.
Methods
To build their models, the research team compiled a dataset of 166 entries from existing literature. These were split into training (80 %) and validation (20 %) sets to ensure robust model performance. Five ML algorithms were trained using the Weka Data Mining software (version 3.8.6, released November 2024).
Each model was evaluated using a comprehensive set of statistical metrics, including mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and various correlation and efficiency measures (R, R2, WI, KGE, NSE, SMAPE).
The input variables reflected key components of the concrete mix: cement (C), water (W), fine aggregates (FAg), coarse aggregates (CAg), plastic waste (PL), steel fibers (SF), fly ash (FA), volume fraction (Vf), fiber length (FbL), and fiber orientation (FbO). The target variable was compressive strength (Cs). A sensitivity analysis—based on the Hoffman and Gardener formula—was used to measure the relative influence of each input on the final strength values.
Results and Discussion
Among all models, Kstar stood out as the top performer. It delivered the highest accuracy (96.5 %) and the lowest error rate (3.5 %), along with the smallest RMSE (2.0 MPa). The Decision Table (DT) model followed closely, offering a solid balance of accuracy (95.5 %) and simplicity. M5Rules also showed promise with a 93 % accuracy rate, though with a slightly higher margin of error.
Efficiency metrics supported these findings. Kstar and DT had nearly perfect scores for KGE and NSE, indicating a strong alignment between predicted and actual results. By contrast, ElasticNet and XNV lagged behind in both accuracy and consistency, with ElasticNet recording the highest RMSE (9.6 MPa).
The sensitivity analysis revealed which mix components had the most influence on compressive strength:
- Water (71 %)
- Fine aggregates (70 %)
- Volume fraction of steel fibers (67 %)
- Fiber orientation (61 %)
- Coarse aggregates (60 %)
In contrast, steel fiber content and fiber length had the least individual influence—just 5 % each—though their effects are still relevant in combination with other factors. The findings emphasize how both the quantity and orientation of fibers can significantly affect a concrete mix’s mechanical performance.
Conclusion
The study makes a compelling case for integrating ML into sustainable construction practices. Kstar, in particular, proved to be a highly accurate and efficient model for predicting compressive strength, with near-perfect correlation values (R = 0.995, R2 = 0.99) and excellent agreement metrics (NSE = 0.99, WI = 1, KGE = 0.985).
While DT offered slightly lower accuracy, it required less computational power, making it a practical option for applications where simplicity and speed are priorities.
Ultimately, using ML to model the strength of concrete made from industrial waste materials offers clear advantages: reduced testing time, lower costs, and improved sustainability. As the construction industry faces increasing pressure to reduce its carbon footprint, data-driven approaches like these could play a crucial role in scaling the use of recycled materials without compromising quality.
Journal Reference
Onyelowe, K. C. et al. (2025). Evaluating the strength of industrial wastesbased concrete reinforced with steel fiber using advanced machine learning. Scientific Reports, 15(1). DOI: 10.1038/s41598-025-92194-3, https://www.nature.com/articles/s41598-025-92194-3
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.