A recent study has shown that machine learning models can accurately predict the compressive strength of carbonated recycled concrete, a discovery that could genuinely help us build better, faster, and greener.
Study: Compressive strength prediction of carbonated recycled aggregate concrete using regression based machine learning models. Image Credit: MeganAnderson/Shutterstock.com
Published in Scientific Reports, the study looked at how well machine learning could anticipate the performance of carbonated recycled aggregate concrete (CRAC). The aim was to move beyond trial and error in mix design and toward something a little more precise. Data-driven concrete, if you will.
At the heart of this research is a clever use of algorithms to help make recycled materials not just usable, but reliably strong. The results support a growing movement in construction: less waste, more reuse, and concrete that’s not just strong, but sustainably so.
A Quick Word on Carbonated Concrete
Let’s briefly set the scene.
Recycled aggregate concrete (RAC) is made using materials from demolished structures, which is a more eco-friendly alternative to digging up fresh stone. Carbonation happens when carbon dioxide reacts with compounds in the concrete, changing its internal structure. Oddly enough, in the case of recycled aggregates, this reaction can actually improve strength and durability.
Understanding just how carbonation changes things is essential if you want to build something that lasts and doesn’t fall down unexpectedly, which is generally the goal.
The Models: More Than Just Math
The researchers tested several regression-based machine learning models, from the straightforward (multilinear regression) to the more sophisticated (random forests and LightGBM). They worked with a dataset built from real-world experimental results, variables like cement ratios, water content, aggregate quality, and how long the concrete had been left to carbonate.
To avoid statistical confusion (and the dreaded multicollinearity), the team devised composite indices, helpful shorthand for complex variables. These included a “mix proportioning index” and an “aggregate performance index,” which sound suitably serious and did their job nicely.
The dataset was split into training and test sets because, as with people, you don’t want to evaluate someone based only on what you’ve already taught them.
What the Data Revealed
Here’s where it gets interesting. The models showed that carbonation duration and aggregate properties are key to strength. LightGBM and decision trees turned in the best scores, with an R2 of 0.991, which, in plain terms, is remarkably accurate.
Still, the random forest model was crowned the most reliable. Not only did it predict strength well, it also revealed which variables mattered most, thanks to SHAP (Shapley Additive Explanations), which sounds vaguely magical but is actually a useful way to interpret machine learning output.
Longer carbonation = stronger concrete.
Notably, the mix proportioning index came out as the top predictor of strength, followed by carbonation degree. Aggregate quality mattered too, but less than you might think. It played a sort of supporting role rather than lead actor.
So What? (A Fair Question)
This isn’t just academic tinkering. These models can be built into design tools, letting engineers create stronger, more sustainable mixes without all the guesswork. They also support the development of building codes that encourage recycled materials, something we’ll need more of if we’re serious about reducing construction’s carbon footprint.
By making concrete performance more predictable, we can reduce waste, speed up design, and cut costs, all without compromising structural integrity. Not bad for something that started out as rubble.
While this study is a meaningful step, there is still room to grow.
Expanding the dataset, testing different curing environments, and experimenting with more advanced algorithms could help improve the models even further. Long-term durability studies and cost analyses will also be key if carbonated recycled concrete is to go mainstream.
In the meantime, it’s reassuring to know that, with the help of machine learning, we’re making even the most unglamorous materials, like old concrete, a little bit smarter.
Journal Reference
Gebremariam, H.G., & et al.(2026). Compressive strength prediction of carbonated recycled aggregate concrete using regression based machine learning models. Sci Rep (2026). DOI: 10.1038/s41598-026-36197-8, https://www.nature.com/articles/s41598-026-36197-8
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.