Prediction model for compressive strength of alkali-activated multi-source solid-waste grouting materials based on generative adversarial networks and machine learning
- 1China Construction Fifth Engineering Bureau Co., Ltd., Changsha, Hunan, China
- 2School of Civil Engineering, Qingdao University of Technology, Qingdao, Shandong, China
- 3Department of Resources and Mechanical Engineering, Lvliang University, Lvliang, Shanxi, China
Abstract
Alkali-activated multi-source solid-waste grouting materials have attracted increasing attention in geotechnical and underground engineering due to their low carbon footprint, environmental friendliness, and potential for large-scale resource utilization. However, the complex composition and multiple influencing factors of these materials pose significant challenges for understanding and predicting their compressive strength, particularly under conditions characterized by small sample sizes, high-dimensional variables, and strong nonlinearity. To address these issues, an experimental database comprising 179 UCS records compiled from three independent experimental studies was established. Fifteen experimental records were reserved for external validation, while WGAN-GP was trained using the remaining 164 model-development samples, thereby keeping the external validation set completely isolated from model development. Multiple machine learning models were then developed to establish a UCS prediction framework. Furthermore, the Newton–Raphson-Based Optimizer (NRBO) was applied to optimize the hyperparameters of the XGBoost model, resulting in an NRBO-XGBoost prediction model. The SHAP method was introduced to interpret the prediction results and quantify the contribution of each input variable. The results demonstrate that WGAN-GP effectively preserves the statistical characteristics and inter-variable correlations of the original data, thereby improving model-training stability and generalization performance. Among all models considered, NRBO-XGBoost exhibited the best predictive performance, achieving an R2 of 0.964, an RMSE of 3.411 MPa, and an MAE of 2.474 MPa in the testing phase. SHAP analysis further revealed that cement content, water-to-binder ratio, curing age, alkali equivalent, and ground granulated blast-furnace slag content play dominant roles in governing compressive strength development. The integrated framework of data augmentation, optimized modeling, and interpretable analysis proposed in this study provides reliable data-driven decision support for mix design optimization and engineering application of alkali-activated multi-source solid-waste grouting materials, and offers a novel technical pathway for performance prediction of complex cementitious materials.
Related
- Machine Learning-Driven Comparative Prediction of Compressive Strength in True Geopolymer and Hybrid Alkali-Activated Cement Systems
- Machine Learning-Empowered Electromagnetic Wave Absorbing Materials: From Forward Prediction to Generative Inverse Design
- Curing regime controlled performance of basalt-modified alkali activated slag: phase composition, porosity profile, microstructure and compressive strength
- GenGLem as a generative framework for capturing chemical cliffs in energetic materials
- Materials Property Prediction with a Site-Resolved Three-Dimensional Voxel Convolution Neural Network
BibTeX
@article{yue2026prediction,
title = {{Prediction model for compressive strength of alkali-activated multi-source solid-waste grouting materials based on generative adversarial networks and machine learning}},
author = {Qiang Yue and Guangming Yu and Yong Yao and Zhiqi Yang and Yufeng Li and Liyang Bai and Guangwei Xiong},
journal = {PLOS One},
year = {2026},
month = sep,
doi = {10.1371/journal.pone.0357051},
url = {https://doi.org/10.1371/journal.pone.0357051},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Prediction model for compressive strength of alkali-activated multi-source solid-waste grouting materials based on generative adversarial networks and machine learning
AU - Qiang Yue
AU - Guangming Yu
AU - Yong Yao
AU - Zhiqi Yang
AU - Yufeng Li
AU - Liyang Bai
AU - Guangwei Xiong
PY - 2026
DA - 2026/09/08
JO - PLOS One
DO - 10.1371/journal.pone.0357051
UR - https://doi.org/10.1371/journal.pone.0357051
AB - Alkali-activated multi-source solid-waste grouting materials have attracted increasing attention in geotechnical and underground engineering due to their low carbon footprint, environmental friendliness, and potential for large-scale resource utilization. However, the complex composition and multiple influencing factors of these materials pose significant challenges for understanding and predicting their compressive strength, particularly under conditions characterized by small sample sizes, high-dimensional variables, and strong nonlinearity. To address these issues, an experimental database comprising 179 UCS records compiled from three independent experimental studies was established. Fifteen experimental records were reserved for external validation, while WGAN-GP was trained using the remaining 164 model-development samples, thereby keeping the external validation set completely isolated from model development. Multiple machine learning models were then developed to establish a UCS prediction framework. Furthermore, the Newton–Raphson-Based Optimizer (NRBO) was applied to optimize the hyperparameters of the XGBoost model, resulting in an NRBO-XGBoost prediction model. The SHAP method was introduced to interpret the prediction results and quantify the contribution of each input variable. The results demonstrate that WGAN-GP effectively preserves the statistical characteristics and inter-variable correlations of the original data, thereby improving model-training stability and generalization performance. Among all models considered, NRBO-XGBoost exhibited the best predictive performance, achieving an R2 of 0.964, an RMSE of 3.411 MPa, and an MAE of 2.474 MPa in the testing phase. SHAP analysis further revealed that cement content, water-to-binder ratio, curing age, alkali equivalent, and ground granulated blast-furnace slag content play dominant roles in governing compressive strength development. The integrated framework of data augmentation, optimized modeling, and interpretable analysis proposed in this study provides reliable data-driven decision support for mix design optimization and engineering application of alkali-activated multi-source solid-waste grouting materials, and offers a novel technical pathway for performance prediction of complex cementitious materials.
SN - 1932-6203
ER -