A data-driven machine learning framework for predicting and optimizing compressive strength of calcium carbide residue-based alkali-activated materials

Calcium carbide residue (CCR) can serve as a waste-derived alkaline component in cementitious binders, but predicting compressive strength across different precursor compositions remains challenging. This study compiled 331 experimental observations at different curing duration and examined seven descriptors: precursor-derived CaO, SiO₂, and Al₂O₃ contributions; CCR-to-binder (C/B), water-to-binder (W/B), curing duration and sand-to-binder (S/B) ratios. Four established regression models were compared using reproducible hyperparameter searches, with standardisation performed within each training fold. Support vector machine (SVM) performed best on the test set, achieving an R² of 0.910, an RMSE of 2.52 MPa, and an MAE of 1.62 MPa. In a separate evaluation using held-out mixture groups, SVM achieved an R² of 0.891, suggesting that the model could predict the strength of mixtures not included in model training. However, further validation using data from independent publications is still needed. SHapley Additive exPlanations (SHAP) analysis identified curing duration and precursor-derived oxide descriptors as influential predictors, although their importance rankings varied across bootstrap model refits. Exploratory model predictions suggested that C/B ratios ranging from 0.03 to 0.21 may be relatively favourable, but this range should not be considered universally optimal. Five independently designed mortar mixtures were then tested at 3, 7, and 28 days. Predictions from the updated model for these 15 age-specific observations yielded an R² value of 0.718. The mixture with a C/B ratio of 0.10 achieved the highest measured 28-day compressive strength of 15.0 MPa. The SEM–EDS and TG results were consistent with differences in matrix development. These findings support the use of data-assisted methods for preliminary mixture screening within the range of materials represented in the dataset.

Authors

Institutions

Publication Details

Journal
Engineering Research Express
Published
2026-11-20
DOI
https://doi.org/10.1088/2631-8695/aec537
Primary Topic
Concrete and Cement Materials Research
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A data-driven machine learning framework for predicting and optimizing compressive strength of calcium carbide residue-based alkali-activated materials

Haozhe Pan, Xingpei Yan
4 links
Engineering Research Express
Concrete and Cement Materials Research
article

A data-driven machine learning framework for predicting and optimizing compressive strength of calcium carbide residue-based alkali-activated materials

Haozhe Pan, Xingpei Yan
article en

Abstract

Calcium carbide residue (CCR) can serve as a waste-derived alkaline component in cementitious binders, but predicting compressive strength across different precursor compositions remains challenging. This study compiled 331 experimental observations at different curing duration and examined seven descriptors: precursor-derived CaO, SiO₂, and Al₂O₃ contributions; CCR-to-binder (C/B), water-to-binder (W/B), curing duration and sand-to-binder (S/B) ratios. Four established regression models were compared using reproducible hyperparameter searches, with standardisation performed within each training fold. Support vector machine (SVM) performed best on the test set, achieving an R² of 0.910, an RMSE of 2.52 MPa, and an MAE of 1.62 MPa. In a separate evaluation using held-out mixture groups, SVM achieved an R² of 0.891, suggesting that the model could predict the strength of mixtures not included in model training. However, further validation using data from independent publications is still needed. SHapley Additive exPlanations (SHAP) analysis identified curing duration and precursor-derived oxide descriptors as influential predictors, although their importance rankings varied across bootstrap model refits. Exploratory model predictions suggested that C/B ratios ranging from 0.03 to 0.21 may be relatively favourable, but this range should not be considered universally optimal. Five independently designed mortar mixtures were then tested at 3, 7, and 28 days. Predictions from the updated model for these 15 age-specific observations yielded an R² value of 0.718. The mixture with a C/B ratio of 0.10 achieved the highest measured 28-day compressive strength of 15.0 MPa. The SEM–EDS and TG results were consistent with differences in matrix development. These findings support the use of data-assisted methods for preliminary mixture screening within the range of materials represented in the dataset.

Engineering Research Express
Jiangsu University Jingjiang College (CN)
Concrete and Cement Materials Research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
A data-driven machine learning framework for predicting and optimizing compressive strength of calcium carbide residue-based alkali-activated materials — Haozhe Pan, Xingpei Yan · Engineering Research Express (2026) | TGRS Research Map | TGRS