Concrete is the most frequently used building material in the world because it offers several advantages over other materials, including integrity, durability, modularity, and cost. Concrete is made up of various elements, including coarse aggregate, fine aggregate, water, and binders. These components are randomly distributed throughout the concrete matrix. In order to effectively evaluate the performance of concrete using advanced design techniques, it is necessary to investigate the mechanical properties of concrete. One of its outstanding mechanical characteristics is its compressive strength. Concrete compressive strength is an important parameter in the design and research of concrete structures, as it is directly related to the safety of the structure and is necessary for determining the performance of structures throughout their life cycle, from new structure design to old structure evaluation. This unique property of concrete can vary depending on several factors, including particle size, water-to-cement ratio, waste composition, and chemical use.1, 2, 3, 4.
Generally, physical experiments are the simplest technique to determine the compressive strength of concrete. Usually cubic or cylindrical specimens are prepared according to a specific planned mixing ratio and cured for the required period of time. The compressive strength can then be measured using a compression testing device. However, this strategy is time-consuming, expensive, and very inefficient. Unlike traditional experimental methods, a specific empirical regression method is provided to estimate the compressive strength of concrete using specific mixing ratios of different components in concrete. Unfortunately, there is a significant nonlinear relationship between concrete mixture and compressive strength, which makes it difficult to obtain a suitable regression equation for this situation. A third approach to capturing concrete behavior is through numerical simulation. Concrete matrix systems pose a major barrier to accurately predicting the compressive strength of concrete materials.5, 6, 7, 8.
With advances in artificial intelligence (AI) in recent years, machine learning (ML) models are widely used in various civil engineering industries, such as building materials, to assist in mix design and performance optimization of cementitious materials. ML is a subset of AI that can be used for various tasks such as classification, regression, and clustering. ML models perform better in large-scale data analysis due to their superior ability to recognize complex and unpredictable relationships between input and output dataset characteristics, making different ML models appropriate for a wide range of data types. Predicting the compressive strength of concrete is just one use of machine learning regression functions. Compared to other traditional regression methods, ML algorithms learn directly from the input data itself and provide highly accurate results on the output data, showing clear advantages over older regression methods.9, 10, 11, 12, 13.
Several successful machine learning models, such as tree-based ensembles, support vector machines, artificial neural networks, and multivariate adaptive regression splines, can simulate a wide range of advanced technical properties of composite materials. Several popular ML models are commonly used on large datasets for compression strength prediction, such as Extreme Gradient Boosting (XGBoost), random forests, and hybrid ML models. ML models are widely used to predict the mechanical properties of concrete. These strategies use large amounts of data to create accurate models. Its predictive accuracy is determined by the data samples utilized in the experimental work during specimen casting or by the literature review. Researchers use these algorithms to predict the mechanical properties of concrete. Several studies have predicted the mechanical properties of a number of newly developed advanced concrete types, including self-healing concrete, recycled aggregate concrete, recycled aggregate concrete with fiber-reinforced rubber, high-strength concrete, and ultra-high-strength concrete. Javed et al. Predicting the compressive strength of sugarcane bagasse ash concrete using gene expression programming. The authors used experimental tests to calibrate and validate the model14,15,16,17,18,19,20,21,22,23,24,25,26.
A major challenge in cement production is that cement contributes significantly to greenhouse gas emissions. In order to support the principles of sustainability and circular economy, recent research has considered environmentally friendly concrete production alternatives. Research highlights the potential for incorporating oil palm by-products into sustainable lightweight structural concrete (SLSC) to offer benefits in life cycle performance, land use efficiency, and cost-effectiveness. A study used advanced machine learning techniques to develop a predictive model for the compressive strength of SLSCs based on a dataset of 449 experimental samples. Among them, the MARS model showed excellent prediction accuracy. The main influencing factors were identified as gravel content, oil palm waste aggregate, and water to binder ratio. This study also includes Monte Carlo simulations and parametric analyzes to verify and enhance the reliability of the proposed model, demonstrating the potential of ML-based approaches in strength prediction in sustainable concrete design.27.
Advanced machine learning approaches are being investigated to improve the prediction of concrete compressive strength, a key factor in structural design and durability assessment. One notable study introduced a hybrid ensemble model (HENSM) that combines multiple traditional machine learning techniques (ANN, linear and nonlinear MARS, GPR, MPMR) with an additional ANN layer to increase prediction accuracy. This study demonstrated that an ensemble model performed better than individual models in both the training and testing phases, based on a 1,030-entry dataset from the UCI Machine Learning Repository. The results suggest that such a hybrid method not only improves prediction accuracy but also helps alleviate overfitting, making it a promising tool for sustainable concrete design and performance prediction.28.
Nondestructive testing (NDT) techniques such as ultrasonic pulse rate and Schmidt rebound hammer tests are widely used due to their simplicity, speed, and efficiency. However, these methods produce highly variable results that can deviate significantly from the actual compressive strength values. To address this limitation, previous studies have considered using artificial neural networks (ANNs) to improve the accuracy of intensity prediction. By training an ANN based on experimental data from the NDT method, the researchers showed that these models closely matched the true compressive strength values. One such study demonstrated the robustness and reliability of an ANN for this purpose and also provided model parameters for practical implementation, such as a spreadsheet for broader accessibility.29.
A recent study investigated the effect of different sand grades on the compressive strength of cement grouts modified with water-reducing polymers according to ASTM and BS standards. In this study, five types of sand and polymer dosages up to 0.16% of the cement weight were investigated and significant improvements were observed in the properties of both fresh and hardened grouts. The water to cement ratio was reduced by up to 54.1% without compromising workability. In particular, coarse sand provides higher strength at low w/c ratios, while finer sands perform better at high w/c ratios. The addition of polymers increased compressive and cylindrical strength by more than 100% in some cases due to gel formation that reduced voids and increased density. Mixtures based on British standards showed up to 71% higher compressive strength than those manufactured according to ASTM specifications, and the finer sand consistently delivered superior bending performance.30.
To address the environmental burden of non-biodegradable polymer waste, recent research has focused on the production of sustainable bricks incorporating cement, fly ash, M-sand, and recycled polypropylene fibers. In this study, advanced machine learning models (ANN, SVM, Random Forest, and AdaBoost) were used to predict the compressive strength of these eco-friendly bricks. The authors enhanced the transparency of the model by using SHAP (SHapley Additive exPlanations) and identified key input variables such as age and fly ash as important predictors. Among the models tested, ANN and Random Forest achieved the highest accuracy, with ANN achieving R² values above 0.99 and the lowest RMSE. This study demonstrates both the effectiveness of machine learning in predicting material properties and the value of explainable AI in sustainable construction.31.
Kumar et al. We focus on predicting the compressive strength of ultra-high performance concrete (UHPC) using advanced machine learning techniques to overcome the limitations of traditional statistical approaches when dealing with nonlinear relationships. A dataset containing 15 input variables related to mixture composition and aggregate properties was used to train several models, including group methods of data processing, recurrent neural networks, LSTM, and Bi-LSTM. Among these, the Bi-LSTM model achieved the best performance during testing with R² 0.9464 and RMSE 0.0482. The results highlight the potential of the model to optimize material selection, reduce experimental effort, and reduce development costs in UHPC mix design.32.
Also, Satyanarayana et al. The seismic performance of reinforced concrete T-beam bridges with elastomeric bearings is investigated. The analysis incorporates region-specific ground motions to assess seismic vulnerability and uses fragility curves as the primary tool to quantify failure probability across different load intensities. To address uncertainty and improve prediction accuracy, this study applies artificial neural networks (ANN) and long short-term memory (LSTM) models to relate structural properties to vulnerability parameters. The findings support the development of accurate fragility curves and enhanced risk assessment of bridge structures during earthquake events.33.
The research structure is organized as follows: Methodology indicating the purpose of the study and the ML model used. Data describing the datasets used in the study. Statistical analysis of data. Machine learning models that provide an overview of each model used. Discussion including results and analysis of results.
