Compressive Strength Prediction of CO2-Cured Low-Strength Concrete Utilising Machine Learning Techniques

The compressive strength (CS) of concrete is an important property in construction, and accurate prediction can help to reduce laboratory testing time and lower experimental costs. Although many studies have been performed to predict the CS of normal concrete, CS prediction for CO₂-cured low-strength concrete is rare. Low-strength concrete is well suited to CO₂ curing because its porous structure allows CO₂ to penetrate more easily. This helps improve early strength while also storing CO₂ within the concrete. Therefore, this study aims to predict the compressive strength of low-strength concrete exposed to CO₂ curing using machine learning models based on mix design parameters and curing conditions. A total of 94 experimental data points were extracted from the literature. Several machine learning (ML) techniques, such as XGBoost, Gradient Boosting, Extra Trees, SVR, and Random Forest, were applied to study how these variables affect the compressive strength of CO₂-cured concrete. At first, ML model evaluation was carried out using an 80% training and 20% testing dataset, followed by a 5-fold cross-validation technique to achieve accuracy. The cross-validation technique is used due to the lower amount of data in the dataset. Based on the results, Extra Trees reached the highest results in the train-test evaluation process, while XGBoost achieved the highest cross-validation. Gradient Boosting also remained a viable alternative, whereas SVR and Random Forest models performed below the leading models. This study contributes to the development of machine learning-based CO₂-cured concrete production by supporting efficient experimental planning, minimizing trial and error, and providing useful data guidance for concrete mix design under CO₂-curing conditions. Although published experimental data on CO₂-cured concrete remain limited, this framework provides an early-stage machine learning based on compressive strength prediction and establishes a foundation for future research.