Compressive strength prediction of machine-made sand concrete based on a Bayesian Optimization-Stacking model

The compressive strength of MSC is highly variable due to the distinctive nature of MS in comparison to conventional sand. In order to enhance the prediction accuracy and dependability of compressive strength on the condition of small-sample data, a predictive model integrating Bayesian optimization with stacked ensemble learning is proposed. Initially, Bayesian optimization was utilized to optimize the hyperparameters of the base learners. Following a comparison with alternative ensemble algorithms, the combination of RF and XGBoost was selected as the base learners. And the LR model was employed as the meta-learner. Secondly, the proposed model was trained and evaluated using 206 authoritative samples. The results demonstrate that the Bayesian optimization-based stacking model markedly enhances the prediction accuracy and exhibits exemplary generalization performance on novel datasets. The main parameter influence on the compressive strength is investigated, which reveals the law of first rising and then falling in relation to the increasing stone powder content, water-cement ratio and sand ratio. The SHAP value analysis indicates that sand ratio and water-cement ratio are the pivotal factors influencing the MSC compressive strength. Owing to the distinctive hierarchical architecture, the model could automatically identify the optimal hyperparameters for base learners through Bayesian optimization and effectively mitigate overfitting on small-sample data. The proposed model may serve as a reference for the optimization of MSC mix designs.