The DOI matches its Crossref record; 189 of the 189 references it cites were found in a registry or an index; 13 of the 44 sources its claims cite were checked for support.
Machine learning in concrete science: applications, challenges, and best practices
Open accessAbstract
Concrete, as the most widely used construction material, is inextricably connected with human development. Despite conceptual and methodological progress in concrete science, concrete formulation for target properties remains a challenging task due to the ever-increasing complexity of cementitious systems. With the ability to tackle complex tasks autonomously, machine learning (ML) has demonstrated its transformative potential in concrete research. Given the rapid adoption of ML for concrete mixture design, there is a need to understand methodological limitations and formulate best practices in this emerging computational field. Here, we review the areas in which ML has positively impacted concrete science, followed by a comprehensive discussion of the implementation, application, and interpretation of ML algorithms. We conclude by outlining future directions for the concrete community to fully exploit the capabilities of ML models.
Overview
CoChat Summary
TL;DR
This review surveys machine learning in concrete science, summarizing 389 peer-reviewed publications and arguing that ML has transformative potential for property prediction, characterization, and simulation but is limited by data sparsity, biased or unrepresentative data, weak validation, and poor interpretability. It proposes best practices around data/tool sharing, linking laboratory and field data, starting with simple models, and knowing when to trust a model.
Key findings
- The literature survey covered 389 peer-reviewed publications; the earliest concrete-and-ML article identified was from 1992, with publication numbers not reaching double digits until 2009. Most studies addressed regression tasks (88.2%), while classification accounted for 10.5%; neural networks were the most commonly applied algorithms.
- Regression datasets were small and skewed: sample sizes ranged from 5 to 262,569, with a median of 174 before preprocessing and 169 after; over 55% of studies had fewer than 200 samples, and only about 11% had more than 1,000 samples. Most regression studies used empirical experimental data (93.6%), with fewer using computational simulation data (7.3%); among experimental studies, 93.8% used laboratory tests and 8.4% used field tests.
- Major application areas include property prediction (e.g., compressive, flexural, tensile, shear strength, elastic modulus, flowability, durability), materials characterization (e.g., nondestructive testing, image segmentation from X-ray computed tomography, crack detection), and materials simulation (e.g., ML potentials for large-scale molecular dynamics of C-S-H with 3,084 atoms for 2 ns).
- The paper highlights data challenges (sparsity, high dimensionality, representation/measurement/temporal/deployment bias), validation challenges (hold-out and cross-validation pitfalls, especially with small datasets), and interpretability challenges (permutation-based diagnostics can generate unrealistic concrete mixtures; ML models identify association, not causation). It also reports that replacing 10% of laboratory training data with field data reduced root mean square error by 43% compared with using only laboratory data in one cited study.
Methods
- This is a review and bibliographic analysis, not a primary experimental study. The authors conducted a literature survey of 389 peer-reviewed publications on ML applications in concrete science, with survey criteria in Supplementary Note 1 and compiled data in Supplementary Dataset.
- Bibliographic data came from Web of Science for the period 1990–2020. The paper also provides a high-level summary of applications, publicly accessible datasets, and current research directions, and discusses challenges and best practices.
- The review synthesizes examples across property prediction, materials characterization, and materials simulation, and discusses validation approaches including two-way and three-way hold-out, k-fold cross-validation, and n × k nested cross-validation.
Limitations
- The paper states that its arguments are “not exhaustive” and aims to initiate constructive discussion rather than provide a complete treatment.
- The bibliographic analysis is limited to the surveyed publications and Web of Science period 1990–2020; detailed survey criteria and compiled data are placed in supplementary materials rather than fully in the main text.
- The review repeatedly notes limitations in the underlying ML literature: small datasets, data bias, inconsistent reporting, validation methods that may be ill-suited to small concrete datasets, and models that are not designed to establish causality.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
11 Valid · 5 Not valid · 19 Not checked · 52 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Failure to assemble a sufficiently large dataset raises the risk of overfitting 82 (Fig. 5).”
§ Machine learning in concrete science: applications, challenges, and best practices > Data sparsity.
What the sources say
- Supportedcites [82] Kassraian-Fard, P., Matthis, C., Balsters, J. H., Maathuis, M. H. & Wenderoth, N. Promises, pitfalls, and basic guidelines for applying machine learning classifiers to psychiatric imaging data, with autism as an example. Front. Psychiatry 7, 177 (2016). source
The 'small sample size' section states directly that when features strongly exceed the number of data samples, 'a first problem is the high risk of overfitting,' and that a small but possibly complex data set can evoke an idiosyncratic fit with poor generalizability. The 'Feature Calculation' section reinforces this, noting that having many predictors relative to data samples causes classifiers to adapt to peculiarities of a specific dataset ('overfitting'), and that dimensionality reduction to alleviate overfitting is crucial 'especially in settings where restricted amount of data are available.' The Summary/Concluding Remarks likewise tie the high-dimensionality of a relatively small data set to easy overfitting. (LLM, conf=0.90)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“If the sample size is relatively small (as is generally the case in practical scenarios), the input variables of a dataset can lose their discriminating ability, as all samples in the dataset seem very distinct from each other with high dimensionality 85 .”
§ Machine learning in concrete science: applications, challenges, and best practices > High dimensionality.
What the sources say
- Partly supportedcites [85] Pestov, V. An axiomatic approach to intrinsic dimension of a dataset. Neural Netw. 21, 204–213 (2008). sourceExperimental — accuracy not yet measured
Pestov (2008) does support the substance of the paraphrase: Sect. 2.2 (p. 2) defines the curse of dimensionality as the situation where 'important features of a dataset sharply concentrate near their median (or mean) values and thus become non-discriminating', explicitly bundling in 'sparseness of points (the distance to the nearest neighbour is comparable to the average distance between two points)', i.e. samples looking mutually distinct in high dimension; Sect. 2.4 (p. 4) adds that large-margin classification with 1-Lipschitz features becomes impossible. The part the source does not support is the small-sample-size condition as the driver: in Sect. 6 (p. 8) realistic/'manageable' sample sizes are tied to the collapse and instability of the separation dimension of a noise-corrupted sample (Example 14), and the proof of Proposition 13 (p. 8) states the opposite of feature non-discrimination for such small samples — a typical high-dimensional sample forms a discrete metric space of diameter ~sqrt(2) whose Lip_1(X) 'will contain numerous 1-Lipschitz functions that are highly non-constant'. The paper's non-discrimination result is asserted for mm-spaces/domains of high intrinsic dimension (Corollary 9, p. 7, requires n >= n0 training points), not for datasets because their sample size is small. (LLM, conf=0.72)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Hence, trust in ML models is strongly related to their interpretability, which has thus become a focus of active research in the ML community 116,132 .”
§ Machine learning in concrete science: applications, challenges, and best practices > Starting with simple models
What the sources say
- Partly supportedcites [116] Lipton, Z. C. The mythos of model interpretability: In machine learning, the concept of interpretability is both important and slippery. Queue 16, 1–28 (2018). sourceExperimental — accuracy not yet measured
The Introduction (p. 1) says many papers propose interpretability as a means to engender trust, and §2.1 (p. 3) says some suggest it is a prerequisite. Lipton questions what trust means and does not establish a strong relationship between trust and interpretability; the claim states that relationship more definitively than the source does. (LLM, conf=0.84)
- Not checked yetcites [132] Schmidt, P. & Biessmann, F. Quantifying interpretability and trust in machine learning systems. Preprint at http://arxiv.org/abs/1901.08558 (2019).
boostedstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In this regard, the knowledge shared in interdependent properties of the same material or in the same property of similar concrete types can be transferred to accelerate the training process and improve the accuracy performance of ML models, even with relatively small amounts of available data 143,144 .”
§ Machine learning in concrete science: applications, challenges, and best practices > CONCLUSIONS AND OUTLOOK
What the sources say
- Partly supportedcites [143] Chen, C., Ye, W., Zuo, Y., Zheng, C. & Ong, S. P. Graph networks as a universal machine learning framework for molecules and crystals. Chem. Mater. 31, 3564–3572 (2019). sourceExperimental — accuracy not yet measured
Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
- Not checked yetcites [144] Ravinder et al. Artificial intelligence and machine learning in glass science and technology: 21 challenges for the 21st century. Int. J App Glass Sci. 12, 277–292 (2021). source
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“However, it is important to note that ML models are not designed to identify causal relationships but rather association relationships 116 .”
§ Machine learning in concrete science: applications, challenges, and best practices > Causality.
What the sources say
- Supportedcites [116] Lipton, Z. C. The mythos of model interpretability: In machine learning, the concept of interpretability is both important and slippery. Queue 16, 1–28 (2018). source
Section 2.2 'Causality' (p. 3) states directly that supervised learning models are 'only optimized directly to make associations' and that 'the associations learned by supervised learning algorithms are not guaranteed to reflect causal relationships' because unobserved common causes may exist. That covers both halves of the paraphrase (association, not causal identification). Lipton adds the caveat that researchers nonetheless use such models to generate hypotheses later tested experimentally, but this does not contradict the claim about what the models are built/optimized to do. (LLM, conf=0.85)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
44 more claims · 3 with their cited sources checked
Create a free account to see all 87 claims and to verify the 31 references still open.
- It may state two separate claims as one. 27
- Probably not a substantial claim. 12
- The quote only partly states this claim. 2
Citations
Cites
Breiman, L. (2001) · Machine Learning · doi:10.1023/a:1010933404324 · Cited by 133,538
Pan, S. J., & Yang, Q. (2010) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2009.191 · Cited by 23,387
Commentary: The Materials Project: A materials genome approach to accelerating materials innovation
Jain, A., Ong, S. P., Hautier, G., Chen, W., Richards, W. D., Dacek, S., Cholia, S., Gunter, D., Skinner, D., Ceder, G., et al. (2013) · APL Materials · doi:10.1063/1.4812323 · Cited by 13,094
Physics-informed machine learning
Karniadakis, G. E., Kevrekidis, I. G., Lu, L., Perdikaris, P., Wang, S., & Yang, L. (2021) · Nature Reviews Physics · doi:10.1038/s42254-021-00314-5 · Cited by 8,718
A Survey on Bias and Fairness in Machine Learning
Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2022) · ACM Computing Surveys · doi:10.1145/3457607 · Cited by 5,812
Lipton, Z. C. (2018) · Queue · doi:10.1145/3236386.3241340 · Cited by 5,427
Adaptive Control Processes: A Guided Tour
Bellman, R. E. (1961) · doi:10.1515/9781400874668 · Cited by 2,802
Recent advances and applications of machine learning in solid-state materials science
Schmidt, J., Marques, M. R. G., Botti, S., & Marques, M. A. L. (2019) · npj Computational Materials · doi:10.1038/s41524-019-0221-0 · Cited by 2,607
Bias in error estimation when using cross-validation for model selection
Varma, S., & Simon, R. (2006) · BMC Bioinformatics · doi:10.1186/1471-2105-7-91 · Cited by 2,485
Machine learning algorithm validation with a limited sample size
Vabalas, A., Gowen, E., Poliakoff, E., & Casson, A. J. (2019) · PLOS ONE · doi:10.1371/journal.pone.0224365 · Cited by 1,901
Research works printed but not linked: 28
- Mehta, P. K. & Monteiro, P. J. M. Concrete: microstructure, properties, and materials (McGraw-Hill Education, 2014).
- Garboczi, E. J., Bentz, D. P. & Frohnsdorff, G. J. The past, present, and future of the computational materials science of concrete. In J. Francis Young Symposium (Materials Science of Concrete Workshop, 2000).
- El-Chabib, H. & Nehdi, M. Effect of mixture design parameters on segregation of self-consolidating concrete. ACI Mater. J. 103, 374–383 (2006).
- Gunasekera, C. et al. Design of Alkali-Activated Slag-Fly Ash Concrete Mixtures Using Machine Learning. ACI Mater. J. 117, 263–278 (2020).
- Pratt, D. & Sansalone, M. Impact-echo signal interpretation using artificial intelligence. ACI Mater. J. 89, 178–187 (1992).
- Li, Z. & Radlińska, A. Artificial intelligence in concrete materials: A scientometric view. In Naser, M. Z. (ed.) Leveraging Artificial Intelligence in Engineering, Management, and Safety of Infrastructure (CRC Press, 2022).
- Rafiei, M. H., Khushefati, W. H., Demirboga, R. & Adeli, H. Neural network, machine learning, and evolutionary approaches for concrete material characterization. ACI Mater. J. 113, 781–789 (2016).
- Rafiei, M. H., Khushefati, W. H., Demirboga, R. & Adeli, H. Novel approach for concrete mixture design using neural dynamics model and virtual lab concept. ACI Mater. J. 114, 117–127 (2017).
- Park, K. B., Jee, N. Y., Yoon, I. S. & Lee, H. S. Prediction of temperature distribution in high-strength concrete using hydration model. ACI Mater. J. 105, 180–186 (2008).
- Kutner, M. H. et al. Applied linear statistical models (McGraw-Hill Irwin, 2005).
- Bažant, Z. R. & Li, G. H. Unbiased statistical comparison of creep and shrinkage prediction models. ACI Mater. J. 105, 610–621 (2008).
- Hubler, M. H., Wendner, R. & Bažant, Z. P. Comprehensive database for concrete creep and shrinkage: Analysis and recommendations for testing and recording. ACI Mater. J. 112, 547–558 (2015).
- Fisher, A., Rudin, C. & Dominici, F. All models are wrong, but many are useful: Learning a variable’s importance by studying an entire class of prediction models simultaneously. J. Mach. Learn. Res. 20, 1–81 (2019).
- Lundberg, S. M. & Lee, S.-I. A Unified Approach to Interpreting Model Predictions. In Guyon, I. et al. (eds.) Advances in Neural Information Processing Systems, vol. 30 (Curran Associates, Inc., 2017).
- El-Chabib, H., Nehdi, M. & Sonebi, M. Artificial intelligence model for flowable concrete mixtures used in underwater construction and repair. ACI Mater. J. 100, 165–173 (2003).
- Basma, A. A., Barakat, S. A. & Al-Oraimi, S. Prediction of cement degree of hydration using artificial neural networks. ACI Mater. J. 96, 167–172 (1999).
- Xi, X., Yang, S., Hu, X. & Li, C. Q. Computational determination of interface fracture properties for concrete by inverse analysis. ACI Mater. J. 117, 43–53 (2020).
- Nehdi, M. L. & Soliman, A. M. Artificial intelligence model for early-age autogenous shrinkage of concrete. ACI Mater. J. 109, 353–362 (2012).
- Aizawa, T. & Yokoshita, T. Applications of neural network technology to the operation and control of cement plants. ZKG Int. 48, 532–538 (1995).
- Schmidt, D. & Schmidt, D. Online prediction of the free lime content in the sintering zone and the use of neural networks for process optimization. ZKG Int. 54, 471–479 (2001).
- Oh, J. W., Lee, I. W., Kim, J. T. & Lee, G. W. Application of neural networks for proportioning of concrete mixes. ACI Mater. J. 96, 61–67 (1999).
- Ji, T., Yang, Y., Fu, M. Y., Chen, B. C. & Wu, H. C. Optimum design of reactive powder concrete mixture proportion based on artificial neural and harmony search algorithm. ACI Mater. J. 114, 41–47 (2017).
- Haj-Ali, R. M., Kurtis, K. E. & Sthapit, A. R. Neural network modeling of concrete expansion during long-term sulfate exposure. ACI Mater. J. 98, 36–43 (2001).
- Chang, C. H. & Yang, C. C. Artificial neural networks in prediction of concrete strength reduction due to high temperature. ACI Mater. J. 103, 68–69 (2006).
- Özbay, E. & Lachemi, M. Relative compressive strength of concretes under elevated temperatures. ACI Mater. J. 109, 165–175 (2012).
- Ünlü, R. An assessment of machine learning models for slump flow and examining redundant features. Comput. Concr. 25, 565–574 (2020).
- Naser, M. Z. & Salehi, H. Machine learning-driven assessment of fire-induced concrete spalling of columns. ACI Mater. J. 117, 7–16 (2020).
- Knaack, A. M., Kurama, Y. C. & Kirkner, D. J. Compressive Strength Relationships for Concrete under Elevated Temperatures. ACI Mater. J. 108, 270–280 (2011).
Cited by
Machine Learning for Perovskite Solar Cells and Component Materials: Key Technologies and Prospects
Liu, Y., Tan, X., Liang, J., Han, H., Xiang, P., & Yan, W. (2023) · Advanced Functional Materials · doi:10.1002/adfm.202214271 · Cited by 174
Alkayem, N. F., Shen, L., Mayya, A., Asteris, P. G., Fu, R., Di Luzio, G., Strauss, A., & Cao, M. (2024) · Journal of Building Engineering · doi:10.1016/j.jobe.2023.108369 · Cited by 164
Modelling of 3D concrete printing process: A perspective on material and structural simulations
Nguyen‐Van, V., Li, S., Liu, J., Nguyen, K., & Tran, P. (2022) · Additive manufacturing · doi:10.1016/j.addma.2022.103333 · Cited by 178
Unboxing machine learning models for concrete strength prediction using XAI
Elhishi, S., Elashry, A., & El‐Metwally, S. (2023) · Scientific Reports · doi:10.1038/s41598-023-47169-7 · Cited by 136
Low clinker systems - Towards a rational use of SCMs for optimal performance
Haha, M. B., Termkhajornkit, P., Ouzia, A., Uppalapati, S., & Huet, B. (2023) · Cement and Concrete Research · doi:10.1016/j.cemconres.2023.107312 · Cited by 131
Metadata
- Authors
- Li, Z., Yoon, J., Zhang, R., Rajabipour, F., Srubar III, W. V., Dabo, I., & Radlińska, A.
- Affiliations
- Pennsylvania State University (US)
- Corresponding author
- Li, Zhanzhao
- Venue
- npj Computational Materials
- Published
- 2022-06-06
- Volume
- 8
- Issue
- 1
- ISSN
- 2057-3960 (ISSN-L)
- OpenAlex
- W4281859794
- Semantic Scholar
- 0ee7b5420699e299a1c55968d6fd67021fc50753
- Cited by
- 388
- References
- 184
- Citations by year
- 2026: 106 · 2025: 113 · 2024: 92 · 2023: 68 · 2022: 9
- FWCI
- 19.80
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- npj Computational Materials (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Massachusetts Institute of Technology
- Topics
- Physical Sciences › Materials Science › Materials Chemistry › Machine Learning in Materials Science
- Physical Sciences › Engineering › Civil and Structural Engineering › Concrete and Cement Materials Research
- Physical Sciences › Engineering › Civil and Structural Engineering › Innovative concrete reinforcement materials
- Keywords
- machine learning, concrete science, concrete mix design, cementitious systems, concrete formulation, best practices
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-04, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-04 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Machine learning in concrete science: applications, challenges, and best practices. https://app.cochat.ai/papers/10.1038/s41524-022-00810-x
- The paper
- Li, Z., Yoon, J., Zhang, R., Rajabipour, F., Srubar III, W. V., Dabo, I., & Radlińska, A. (2022). Machine learning in concrete science: applications, challenges, and best practices. npj Computational Materials. https://doi.org/10.1038/s41524-022-00810-x
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.