Valorization of waste glass into sustainable cementitious materials: An intelligent approach for fresh, mechanical, and durability performance assessment

To valorize waste glass into sustainable cementitious materials as a promising cement alternative in concrete production, published studies have accumulated extensive data on glass powder concrete (GPC) performance through tedious experiments. In this study, an intelligent approach for the comprehensive performance assessment of GPC was proposed using exceeding 1800 sets of data. 22 optimization algorithms were assessed, among which the superior algorithm was employed to develop a hybrid machine learning (ML) model for the assessment of GPC with various mix proportions in terms of slump (SL), compressive strength (CS), and total charge passed (TCP). The results indicated that the hybrid ML algorithm SSA-XGB provided the highest prediction accuracy, with R² values of 0.9411, 0.9330, and 0.9449 for SL, CS, and TCP, respectively. Additionally, following the incorporation of GP particle size into the model, an improvement of 1.51% in the R² value of SSA-XGB was observed. An interpretability analysis was performed to elucidate the influence of input features on the performance of GPC, and it was demonstrated that GP contents ranging from 25 to 160 kg/m³ were associated with superior performance enhancements. Finally, an interactive program was developed to enable batch processing and automatic prediction of the comprehensive performance of GPC without the necessity for labor-intensive experiments. This study provides an innovative and intelligent approach for the comprehensive performance assessment of GPC, contributing to the efficient resource utilization of waste glass in the construction sector. • An intelligent approach was proposed to assess the performance of glass powder concrete. • Among the 22 optimization algorithms, SSA, AVOA, and GTO algorithms demonstrate excellent optimization performance. • SSA-XGB algorithm stands out with the highest prediction accuracy. • The inclusion of the SCM particle size factor contributes to improving the model's accuracy. • A program was developed to achieve batch and automated predictions for convenient use.