The environmental burden associated with conventional cement-based materials has intensified research for sustainable alternatives with lower carbon footprints. For this, gypsum-based composites reinforced with agricultural waste, such as wheat straw, offer a promising solution. However, their mechanical performance is governed by nonlinear and complex interactions among multiple mixture parameters. This study proposes a comprehensive machine learning (ML) framework to predict the compressive and flexural strength of wheat straw reinforced gypsum composites. A dataset comprising 161 experimental samples was used and five ML models: artificial neural network, Gaussian process regression (GPR), random forest, extreme gradient boosting, and support vector machine, were used. Model performance was assessed using tenfold cross-validation with multiple statistical metrics along with Taylor diagram analysis. Among the evaluated models, GPR demonstrated superior predictive capability for both compressive and flexural strength, while providing uncertainty quantification that enhances reliability for engineering applications. Feature importance and SHapley Additive exPlanations analyses were employed to improve model interpretability, revealing gypsum strength as the most influential parameter, with water-related parameters, wheat straw content, and chemical additives contributing secondary effects. The proposed ML-based framework provides acceptable and interpretable predictions, offering the optimization of sustainable gypsum composites while reducing experimental efforts and supporting environment-friendly construction.
Ahmad et al. (Thu,) studied this question.