Evaluation of the Impact of Different Combinations of Sustainable Concrete Materials on the Compressive Strength of Concrete Using Machine Learning Techniques

Evaluation of the Impact of Different Combinations of Sustainable Concrete Materials on the Compressive Strength of Concrete Using Machine Learning Techniques

Chandra Shekhar Kumar, Sujit Kumar, Dr. Bimal Kumar

Government Engineering College Jamui India

Dr. Sanjay Kumar

National Institute of Technology Patna India

Abstract – Nowadays, the application of Machine Learning (ML) techniques has gained significant popularity across various research domains. Concrete compressive strength (CCS) is a critical parameter for ensuring the long-term safety and durability of reinforced cement concrete (RCC) structures. Concrete is a composite material consisting of cementitious materials, fine aggregates, coarse aggregates, water, and chemical admixtures, and its compressive strength depends on the combination and proportion of these constituents. In this study, a dataset comprising 125 samples was used for the development of ML-based prediction models. The data were collected from laboratory experimental investigations conducted in the laboratory and the relevant published literature. Several ML algorithms, including Linear Regression, Ridge Regression, Lasso Regression, Decision Tree, Random Forest, Gradient Boosting, Support Vector Regression (SVR), XGBoost, and Multi-Layer Perceptron (MLP) regression, were employed to predict the compressive strength of concrete.The results indicate that the MLP regressor achieved the highest coefficient of determination (R²) among all the models considered. It was also observed that the type of concrete materials used, selected input features, target variable, dataset size, and choice of ML algorithm significantly influence model performance. All results and graphical analyses were generated using Python programming on the Google Colab platform. The optimized performance metrics for the MLP regressor include an R² value of 0.839, a root mean square error (RMSE) of 6.656, and a mean absolute error (MAE) of 5.414.

Keywords Concrete Compressive Strength, MLP regression, Gradient Boosting, Coefficient of determination

INTRODUCTION

Concrete is the most widely used construction material in the world due to its durability, versatility, and relatively low cost. One of the key indicators of its performance and structural integrity is compressive strength, which directly influences the safety, serviceability, and longevity of concrete structure Accurate evaluation of concrete compressive

strength is therefore essential during both construction and maintenance phases. A critical aspect of ensuring the structural integrity and service life of concrete structures lies

in accurately evaluating their compressive strength and overall quality. Traditionally, this evaluation has relied on Destructive Testing (DT) methods such as compression tests on concrete cylinders or cores, which, while highly accurate, can be time consuming, costly. To overcome these limitations, Non-Destructive Testing (NDT) technique such as ultrasonic pulse velocity (UPV), rebound hammer, and ML models have gained popularity. These NDT methods allow for in-situ evaluation without damaging the structure, though they often require calibration and interpretation based on empirical correlations. In recent years, the integration of Machine Learning (ML) in civil engineering has introduced new opportunities for predicting concrete properties with high accuracy and efficiency.

The objective is to enhance the reliability, speed, and cost effectiveness of concrete strength and quality assessments in modern construction and maintenance practices.

The work primarily provide a thorough overview of the state of the art in computer vision-based defect detection and condition assessment for civil infrastructure made of asphalt and concrete.

The researchers works are based on without requiring knowledge of the concrete's composition or history, the researchers classified and predicted concrete strength based only on UPV and RH values.

An early attempt to estimate the compressive strength of lightweight foamed concrete using the Extreme Learning Machine, or ELM, a regression-based soft computing model that is an enhanced version of neural network models. A novel crack detection framework is proposed to automatically assess the cracking in concrete post-fire conditions .

The researchers optimized artificial neural networks using metaheuristics to predict the compressive strength of concrete, we saw a significant improvement in test RMSE values of 39% to 48%. Combining NDT tests with various

mix percentage designs and curing ages is an efficient way to forecast the compressive strength of concrete. In particular, back propagation neural network (BPNN) and support vector machine regression (SVR) models are developed .

The machine learning models demonstrated a strong correlation with the test results, achieving a R2 value of 0.84 for the XGBoost, in order to examine the synergistic effects of a cement-based mixture consisting of RHA and FA in various proportions on concrete's fresh, hardened, non- destructive, and microscopic properties.

The long-term effects of mixing different chemical admixtures on concrete's compressive strength using sophisticated experimental data and machine learning models with a degree of accuracy and detail that has been comparatively understudied.

The assessment, repair, and retrofitting of masonry structures are described along with the latest developments in machine learning approaches.

With an emphasis on two main methodsdestructive testing (DT) like core drilling and non-destructive testing by calibration (NDTBC)this study methodically examines the concrete strength testing systems in China and Europe. A comparable database is created by standardized processing using 360 component rebound and 240 core specimens with concrete strength classes ranging from C30 to C70 from eight large datasets (24 small datasets) of four typical Chinese projects .

The effects of using quartz powder (QP) and snail shell powder (SSP) in place of some cement can lessen the environmental impact of concrete. The study evaluates the mechanical, microstructural, fresh, and non-destructive qualities of SSP and QP contents at 2.510% as an additional cementitious material (SCM) in the creation of environmentally friendly concrete.

Overall, the results demonstrate the efficacy of moisture- aware, integrated non-destructive testing techniques for accurate on-site strength evaluation and validate the structural viability of recycled aggregate concrete. These results support the idea that recycled aggregates should be used more widely in sustainable concrete applications.

587 concrete mix samples were gathered from the literature to create a dataset. Seven prediction methods were used: Multi-Layer Perceptron, Linear Regression, Support Vector Regression, Decision Tree Regressor, Random Forest Regressor, Gradient Boosting Regressor,,and Extreme Gradient Boosting. Missing data treatment, outlier detection encoding, and feature normalization for Principal Component Analysis were all part of preprocessing. Coefficient of determination (R2), root mean squared error, mean absolute error, and mean absolute percentage error with cross- validation were used to evaluate performance. Superior

accuracy was demonstrated using Extreme Gradient Boosting (R2 = 0.952 for tensile strength and 0.954 for compressive strength). The mechanical qualities of reinforced concrete structures are increasingly being assessed using machine learning (ML) and image processing (IP). This study focuses on hybrid concrete (HC), which mproves mechanical performance and promotes sustainability by including cement replacement materials (CRM) such fly ash and silica fume. To enhance two-parameter SonReb-based AI models by adding a binary input variable that indicates the specimen geometry type (cube or cylinder) and examines the impact of model architecture by contrasting three distinct AI algorithms: Artificial Neural Networks (ANN), Deep Neural Networks (DNN), and Adaptive Neuro-Fuzzy Inference Systems (ANFIS).

The results encourage the wider use of alkali-activated concrete (AAC) as an environmentally friendly building material and provide practical insights for improving AAC formulations study applied artificial neural networks to predict the compressive strength of concrete.

The model showed better accuracy than traditional regression methods.It demonstrated the potential of ML for nonlinear material behavior prediction.

They developed several data-mining techniques for predicting concrete strength. ANN models outperformed statistical approaches for complex mix compositions. The study emphasized the importance of dataset quality.

METHODOLOGY

This study adopts the results from the experiment work in the laboratory

Concrete samples were prepared according to standard mix design procedures. The curing process was carried out under controlled conditions for 7, 14, and 28 days to assess strength development over time.

• Sample Types: Concrete cubes

• Sample Sizes: Standard

  150 mm x 150 mm x 150 mm cubes

• Curing Periods: 7, 14, and 28 days

Fig. 1 Cube mould for sample preparation

Fig. 2 Vibration table for sample preparation

Fig. 3 Cube sample preparation

Fig. 4 Cube compressive strength test and compression testing machine setup

Slump Test and Result

Slump Test is applicable for measuring the workability of fresh concrete. Slump Value varies based on the various exposure condition and type of the construction work. CCS results were obtained in Compression Testing Machine by testing the standard cubes after proper curing in 7, 14 and 28 Days.

Fig. 5 Slump test of fresh concrete

Fig. 6 Reading duration compression testing of cube sample

Results

Table 1 Model performance comparison

From the table 1 it is found that the coefficient of determination(R2) is minimum in Linear regression model and maximum in MLP Regressor. The two best ML model based on the available data are MLP Regressor and Gradient Boosting based on the value of R2.

Fig. 7 Correlation Heat Map of concrete Features

Fig. 8 Actual vs Predicted Values of CCS

The Actual vs Predicted plot in figure 8 compares the models strength predictions with the real measured CCS values to show how accurately the model

performs.

Points closer to the diagonal line indicate high prediction accuracy, while larger deviations reveal modelling errors or noise in the data.

Fig. 9 Comparison of R2 value

Comparing R² values in figure 9 shows how well different models explain the variability in concrete compressive strength. A higher R² indicates better predictive performance and a stronger fit between the model and actual data.

Fig. 10 Comparison of RMSE value

A lower RMSE value in figure 10 indicates better model performance, as it reflects smaller average prediction errors.

Fig. 11 Concrete Compressive Strength distribution

The distribution in figure 11 shows how concrete compressive strength increases with curing age, reflecting the hydration process and strength gain over time.

Fig. 12 Feature vs CCS Plot

Feature vs CCS plots in figure 12 illustrate how each input variable influences the concretes compressive strength across its range.They help identify positive or negative relationships, nonlinear trends, and key factors that strongly impact strength prediction.

Fig. 13 Box plot Age vs CCS

The box plot in Figure 13 shows how concrete compressive strength varies across different curing ages, highlighting the median strength and spread of values. It helps identify outliers and understand how strength distribution changes as concrete matures over time.

Fig. 14 Sns pair plot features vs Target

A Seaborn (SNS) pair plot visualizes pairwise relationships between features and the target, revealing correlations, trends, and distributions in one view.

CONCLUSION

The results indicate that the performance of prediction models varies across different machine learning (ML) techniques, with each model exhibiting its own advantages and limitations. The effectiveness of an ML model primarily depends on the selected input features, target variable, and the size of the dataset. Among all the ML models considered, the MLP regressor was found to be the most optimal for model development.

For future studies, it is essential to carefully select the most suitable ML model for predicting concrete compressive strength (CCS) based on different concrete material compositions, while also incorporating standard code guidelines to ensure that all significant features are considered. Optimization techniques should be employed to enhance prediction accuracy. The results further reveal that statistical indicators such as R², RMSE, and MAE are key parameters for evaluating ML model performance.

Supplementary cementitious materials (SCMs) represent an emerging and important area of research, as cement production is a major contributor to global carbon dioxide emissions. Materials such as fly ash, silica fume, ground granulated blast furnace slag, rice husk ash, and bagasse ash are increasingly being used as partial replacements for cement in concrete production. Consequently, ML models can be continuously developed and updated to incorporate innovations in concrete materials for accurate prediction of concrete compressive strength. The study demonstrates that machine learning techniques effectively capture the complex nonlinear relationships between sustainable concrete material combinations and compressive strength.

Results indicate that optimized blends of supplementary cementitious and recycled materials can achieve comparable or improved strength performance relative to conventional concrete.

Overall, ML-based evaluation provides a reliable decision- support tool for designing high-performance, sustainable concrete in accordance with material efficiency and strength requirements.

REFERENCES

1. Koch, C., Georgieva, K., Kasireddy, V., Akinci, B., & Fieguth, P. (2015). A review on computer vision based defect detection and condition assessment of concrete and asphalt civil infrastructure. Advanced Engineering Informatics, 29(2), 196 210.

   https://doi.org/10.1016/j.aei.2015.01. 008

2. Amini, K., Jalalpour, M., & Delatte, N. (2016). Advancing concrete strengt prediction using non-destructive testing: Development and verification of a generalizable model. Construction and Building Materials, 102, 762768. https://doi.org/10.1016/j.conbuildmat.

   2015.10.131

3. Yaseen, Z. M., Deo, R. C., Hilal, A., Abd, A. M., Bueno, L. C., Salcedo-Sanz, S., & Nehdi, M. L. (2018). Predicting compressive strength of lightweight foamed concrete using extreme learning machine model. Advances in Engineering Software, 115, 112125. https://doi.org/10.1016/j.advengsoft.2

   017.09.004

4. Andrushia, A. D., Anand, N., Neebha, T. M., Naser, M. Z., & Lublóy,

   E. (2022). Autonomous detection of concrete damage under fire conditions. Automation in Construction, 140, 104364.

   https://doi.org/10.1016/j.autcon.2022. 104364

5. Selçuk, S., & Tang, P. (2023). A metaheuristic-guided machine learning approach for concrete strength prediction with high mix design variability using ultrasonic pulse velocity data. Developments in the Built Environment, 15, 100220.

   https://doi.org/10.1016/j.dibe.2023.100220

6. Zhang, W., Liu, D., & Cao, K. (2024). Prediction of concrete compressive strength using support vector machine regression and nondestructive testing. Case Studies in Construction Materials, 21, e03416.

7. Karim, R., Islam, M. H., Datta, S. D., & Kashem, A. (2024). Synergistic effects of supplementary cementitious materials and compressive strength prediction of concrete using machine learning algorithms with SHAP and PDP analyses. Case Studies in Construction Materials, 20, e02828. https://doi.org/10.1016/j.cscm 2024.e0 2828

8. Heidari, S. I. G., Safehian, M., & Moodi, F. (2024). Predictive modelling of the long-term effects of combined chemical admixtures on concrete compressive strength using machine learning algorithms. Case Studies in Chemical and Environmental Engineering, 10, 101008. https://doi.org/10.1016/j.cscee.2024.1 01008

9. Keshmiry, A., Hassani, S., Dackermann, U., & Li, J. (2024). Assessment, repair, and retrofitting of masonry structures: A comprehensive review. Construction and Building Materials, 442, 137380. https://doi.org/10.1016/j.conbuildmat. 2024.137380

10. Wang, C., Yuan, E., Chang, L., Zhang, S., Chen, W., Li, X., Zhao, Y., & Qiu, F. (2025). Test and evaluation method of compressive strength of existing concrete

    structures. Case Studies in Construction Materials,23,e05233.https://doi.org/10.1016/j.cscm 2025.e052 33

11. Sobuz, M. H. R., Aditto, F. S., Tilak, M. K. C., Shaikh, F. U. A.,

    Rahman, S. A., Kabbo, M. K. I., Jameel, M., & Khan, M. M. H. (2025). Comparative performance evaluation of quartz and snail shell powders modified concrete: Mechanical, machine learning, and microstructural assessments. Journal of Materials Research and Technology, 36, 77667782. https://doi.org/10.1016/j.jmrt.2024.08.229

12. Abdennaji, T. S., Tipu, R. K., & Alassaf, Y. (2025). Predicting compressive and tensile strength of concrete with different sand types using machine learning. Ain Shams Engineering Journal, 16, 103474. https://doi.org/10.1016/j.asej.2024.10 3474

13. Plevris, V., Al-Sayegh, A. T., & Mir, J. (2025). Nondestructive evaluation of hybrid concrete properties using image processing and machine learning. Structures, 79, 109423.

    https://doi.org/10.1016/j.istruc.2024.1 09423

14. Alavi, S. A., & Noel, M. (2025). Effect of model architecture and input parameters to improve performance of artificial intelligence models for estimating concrete strength using SonReb. Engineering Structures, 323, 119285. https://doi.org/10.1016/j.engstruct. 2 024.119285

15. Emarah, D. A. (2025). Multivariate predictive modeling of compressive strength in ground granulated blast furnace slag/fly ashbased alkali-activated concrete. Cleaner Engineering and Technology, 27, 101021. https://doi.org/10.1016/j.clet.2024.101021

16. Yeh, I. C. (1998). Modeling of strength of high-performance concrete using artificial neural networks. Cement and Concrete Research, 28(12), 17971808

17. Yeh, I. C. (2006). Analysis of strength of concrete using data mining techniques. Cement and Concrete Composites, 28(4), 350358.

18. Sobuz, M. H. R., Aditto, F. S., Tilak, M. K. C., Shaikh, F. U. A.,

    Rahman, S. A., Kabbo, M. K. I., Jameel, M., & Khan, M. M. H. (2025). Comparative performance evaluation of quartz and snail shell powders modified concrete: Mechanical, machine learning, and microstructural assessments. Journal of Materials Research and Technology, 36, 77667782. https://doi.org/10.1016/j.jmrt.2024.08

    .229

19. Joshi, T., Mathur, P., Oza, P., Agrawal, S., & Narmawala, H. (2024). Predictive modelling of sustainable concrete compressive strength using advanced machine learning algorithms. Advances in Civil and Architectural Engineering, 15(29), 168192. https://doi.org/10.13167/2024.29.11

20. Amin, M. N., et al. (2025). Machine learningdriven optimization of compressive and tensile strength in concrete with GGBS, eggshell powder, and waste glass powder. Scientific Reports.

21. More, S. S., & Kambekar, A. R. (2025). Performance evaluation of compressive strength of concrete using different machine learning

    algorithms. Challenge Journal of Concrete Research Letters. https://doi.org/10.20528/cjcrl.2025.02.002

22. Onyelowe, K. C., Ebid, A. M., & Ramos Jiménez, R. B. (2025). Modeling the compressive strength of concrete at different curing regimes using machine learning. Discover Sustainability, 6, 358. https://doi.org/10.1007/s43621-025-01195-9

23. Poudel, S., Gautam, B., Bhetuwal, U., et al. (2025). Machine learning- based prediction of compressive strength of sustainable concrete incorporating waste glass as a partial replacement for cement. Preprints.org.

24. Ozuligbo, I. J., Nwadiani, E. V., & Okoh, J. O. (2025). Machine learning-based prediction of compressive strength of recycled aggregate concrete: A comparative study of different algorithms. Nnamdi Azikiwe University Journal of Civil Engineering.

25. Prediction of compressive strength of sustainable foam concrete using individual and ensemble machine learning approaches. (2025). Journal of Construction Materials.

26. Environmentally friendly concrete compressive strength prediction using hybrid machine learning. (2025). Sustainability, 14(20), 12990.

27. Synergistic effects of supplementary cementitious materials and compressive strength prediction using ML with SHAP and PDP analyses. (2023). Case Studies in Construction Materials, 20, e02828.

28. Mouawad, F., Homsi, F., Geara, F., & Mina, R. (2025). Predicting compressive strength of sustainable concrete using machine learning and artificial neural networks. Construction Materials, 5(3), 56.

29. Hammoudi, A., et al. (2021). Machine learning for predicting compressive strength of sustainable cement paste incorporating copper mine tailings as SCM. Case Studies in Construction Materials.

30. Aslam, F., et al. (2021). Supplementary cementitious materials in blended cement concrete: ML evaluation of compressive strength. Journal of Building Engineering.

31. Machine learning approaches for forecasting compressive strength of high strength concrete usin Python. (2025). Scientific Reports.

32. Estimating concrete compressive strength using machine learning models (ANN, SVM, etc.). (2024). Preprints.org.

33. Song, H., Aslam, F., Ahmad, A., & Farooq, F. (2021). Predicting compressive strength of concrete with fly ash using GEP, ANN, and decision tree algorithms. Constr. Build. Mater.