Smarter Concrete with Hybrid Random Forest Strength Predictions
Concrete remains the most widely used construction material on the planet, but it comes at a cost. Traditional cement production accounts for nearly 8% of global CO₂ emissions, sparking the industry’s search for greener alternatives.
One such solution is high-volume fly ash (HVFA) self-compacting concrete (SCC), which replaces a significant portion of cement with industrial by-products. The material not only tackles waste but also improves workability and durability. Yet, predicting its performance has been a complex puzzle, often hindered by the opaque nature of machine learning (ML) models.
Researchers from India’s Muzaffarpur Institute of Technology and Dayananda Sagar College of Engineering, alongside Thailand’s Thammasat University, have set out to change this narrative. Their collaborative study, Smart Prediction: Hybrid Random Forest for High-Volume Fly Ash Self-Compacting Concrete Strength, delves into how advanced modelling techniques can make strength prediction more accurate and transparent.
The challenge of prediction
One of the persistent challenges with HVFA-SCC is its variability. Replacing cement with fly ash and silica fume introduces complex interactions that affect both fresh and hardened properties. Engineers must contend with multiple factors, such as cement content, replacement ratios, curing age, and test parameters such as L-box blocking ratio, J-ring results, and V-funnel time. Traditional predictive models struggle to capture this complexity.
Enter machine learning. Random Forest (RF) models have already shown promise in construction materials research, yet their “black-box” reputation often leaves practitioners wary. Engineers want reliable predictions but also need to understand why models deliver certain outcomes. Bridging this gap is essential for real-world adoption.
A hybrid machine learning approach
The research team tested several hybrid RF models, combining RF with particle swarm optimisation, Bayesian optimisation, and differential evolution (RF-DE). These enhanced models were trained on a range of experimental inputs, including cement, silica fume, fly ash, T-500 time, maximum spread diameter, and curing age.
To validate the accuracy, the team deployed tools such as SHapley Additive exPlanations (SHAP), regression error characteristic curves, and uncertainty analysis. Their findings were clear: the RF-DE model stood out, delivering the most precise predictions of compressive strength (CS) across different mix conditions.
From data to practical tools
The most compelling outcome wasn’t just the improved accuracy. Recognising that engineers often struggle with ML’s lack of transparency, the team built an open-source graphical user interface (GUI) based on the RF-DE model. This practical tool allows engineers to input mix parameters and receive compressive strength predictions, offering a clear pathway to optimise SCC designs without advanced coding or statistical expertise.
As the authors put it: “The user-friendly tool can provide precise CS predictions under various mix conditions, enabling engineers to optimise mix proportions, support sustainable concrete design, and promote practical ML applications in the industry.”
Optimising fly ash and silica fume content
The study went further by exploring how different proportions of fly ash and silica fume (SF) affect performance. It found that the simultaneous addition of FA and SF reduced bleeding and segregation in SCC mixtures. The sweet spot? An SF content of 6–8% yielded the highest compressive strength after 90 days of curing.
Microstructural analysis using scanning electron microscopy (SEM) and X-ray diffraction (XRD) confirmed these results. HVFA-SCC with SF displayed a denser mortar matrix and fewer voids, directly contributing to improved mechanical performance. In other words, the microstructure backed up the data-driven predictions.
Implications for the construction industry
The implications of this research stretch well beyond the laboratory. If engineers can accurately predict the performance of HVFA-SCC, they can:
- Reduce cement consumption and associated CO₂ emissions.
- Repurpose industrial waste products like fly ash and silica fume.
- Improve quality control on construction sites.
- Accelerate the adoption of sustainable concretes in large-scale infrastructure projects.
For industry investors and policymakers, such tools signal a future where material design aligns seamlessly with sustainability targets without sacrificing safety or performance.
Towards more transparent AI in construction
While AI is revolutionising industries, the construction sector often lags in adoption due to concerns over trust and interpretability. This research demonstrates how hybrid ML approaches, combined with accessible user tools, can build that trust. The open-source GUI is a tangible step towards demystifying AI and ensuring it becomes a practical ally rather than a black box.
Suraparb Keawsawasvong, one of the co-authors, emphasised this point: “Our aim was not only to improve accuracy but to provide a practical tool that engineers can trust and use. By combining hybrid models with explainability, we move closer to a standardised framework for sustainable concrete development.”
Building confidence in green concrete
For decades, concrete’s sustainability challenge has loomed large. Yet with research like this, the pathway towards greener, stronger, and smarter concrete becomes clearer. Hybrid RF-DE models, supported by microstructural validation, prove that data-driven methods can unlock new performance standards for HVFA-SCC.
As cities expand and infrastructure needs grow, the demand for sustainable construction materials will only intensify. By equipping engineers with predictive tools that are accurate and transparent, the industry can confidently embrace eco-friendly concretes, turning sustainability into standard practice rather than a niche ambition.
A positive step for sustainable construction
The study’s contribution lies not just in improving predictions but in reshaping how the industry approaches innovation. By blending advanced analytics, explainable AI, and practical user tools, it addresses a critical bottleneck in sustainable concrete adoption.
The message is clear: sustainable concretes are no longer an experimental novelty. With the right tools, they’re ready for widespread, confident use in modern construction.
