Integrating machine learning with experimental data to predict tensile strength of cement pastes with binary cementitious systems
| dc.contributor.author | Khan, Ameer Murad | |
| dc.contributor.author | Hussain, Qudeer | |
| dc.contributor.author | Mohamad, Hisham | |
| dc.contributor.author | Jirasakjamroonsri, Amornthep | |
| dc.contributor.author | Anotaipaiboon, Weerachai | |
| dc.contributor.author | Saingam, Panumas | |
| dc.contributor.author | Phetmanee, Surasak | |
| dc.contributor.author | Thansirichaisree, Phromphat | |
| dc.date.accessioned | 2026-08-06T10:56:05Z | |
| dc.date.available | 2026-08-06T10:56:05Z | |
| dc.date.issued | 2026-07-01 | |
| dc.description.abstract | This study demonstrates a concerted experimental and machine learning model to determine the direct tensile strength of cement pastes with the addition of supplementary cementitious materials (SCM) used as binary cementitious systems. Five types of SCM were used as single binary substitutes at levels of 5–25 % each to form a complete experimental database of 135 cement paste mixtures at water-cement ratios of 0.30–0.40, respectively, to provide a systematic and applied evaluation of the individual contribution of each material to tensile performance. Briquette tests were used to measure direct tensile strength at the binder level to isolate behaviour. Ten mix design parameters were used to train six machine learning regressors (Decision Tree, Random Forest, Gradient Boosting, XGBoost, CatBoost and Support Vector Regression) to predict tensile strength. Support Vector Regression and XGBoost had the best predictive accuracy with R<sup>2</sup> that reached 0.88 and lowest RMSE of 3.3. Analysis of feature importance and partial dependence indicated that the dominant predictors were the water-cement ratio, cement content, and dosage of fly ash and silica fume had a small optimal replacement range, which is in line with the experimental data. The existence of a correspondence between the trends of machine learning and the known hydration mechanisms proves the physical plausibility of the models. The most successful model was an interactive web based tool, TensAile-Lab, to be used in support of fast, data-driven mix design decisions. The suggested framework illustrates the use of interpretable machine learning to forecast tensile behavior with high reliability and speed up the creation of optimized, low-carbon cementitious systems. | |
| dc.identifier.citation | Composites Part C Open Access, 20, 2026 | |
| dc.identifier.doi | 10.1016/j.jcomc.2026.100749 | |
| dc.identifier.issn | 26666820 | |
| dc.identifier.other | 2-s2.0-105040542087 | |
| dc.identifier.uri | https://dspace.kmitl.ac.th/handle/123456789/18243 | |
| dc.source | Composites Part C Open Access | |
| dc.subject | Machine learning (ML) | |
| dc.subject | Model deployment | |
| dc.subject | Supplementary cementitious materials (SCMs) | |
| dc.subject | Support vector regressor (SVR) | |
| dc.subject | Tensile strength (TS) | |
| dc.title | Integrating machine learning with experimental data to predict tensile strength of cement pastes with binary cementitious systems | |
| dc.type | Article |