Machine learning prediction of drying shrinkage for alkali-activated materials and multi-objective optimization

The aim of this study is to perform multi-objective optimization of alkali activated materials (AAMs) through a machine learning approach, focusing on three key factors: drying shrinkage, cost and CO₂ emission. Five machine learning models were trained and evaluated based on 1383 hybrid design data, and the results showed that the gradient boosted regression (GBR) model performed best in terms of prediction accuracy. The main factors affecting the drying shrinkage of AAMs, including curing time, the content of SiO₂ and Na₂O in sodium silicate, and the ratio of fly ash to slag content, were identified through the Shapley Additive Explanations (SHAP) method and sensitivity analysis. Further, a multi-objective optimization model was constructed by combining the GBR model and the Non-dominated Sorting Genetic Algorithm II (NSGA-II) algorithm while considering drying shrinkage, cost and CO₂ emission. The model can help researchers and engineers to quickly design and optimize the ratios of AAMs to meet practical engineering needs.