<p>Accurate prediction of concrete shrinkage is crucial for enhancing durability and reducing service-life issues in construction. Conventional empirical methods frequently fail to account for the intricate nonlinear interactions between material components and ambient variables that affect shrinkage behavior. This study utilized machine learning techniques to improve predictive accuracy, optimizing Support Vector Regression (SVR) and Decision Tree Regression (DTR) models with advanced metaheuristic algorithms, specifically the Arithmetic Optimization Algorithm (AOA) and Aquila Foraging Technique (AFT). This paradigm sought to equilibrate precision, interpretability, and computing efficiency. The collection included unsealed shrinkage measurements affected by mix composition and environmental factors. Recursive Feature Elimination was employed for feature selection, guaranteeing the retention of the most pertinent variables for modeling. The model’s performance was assessed using common measures, such as R² and RMSE, with independent validation and test sets utilized to reduce overfitting. The optimized models exhibited superior predictive capacity, with coefficients of determination surpassing 0.98 in the test data. The dual-tree AOA-based method had the maximum accuracy, but with increased computational expense. The comparison research revealed a trade-off between predictive effectiveness and efficiency, indicating that simpler models may be more beneficial for quick applications. These findings indicate that optimization-assisted machine learning can function as a dependable instrument for predicting shrinkage behavior. The practical consequences encompass aiding mix design decisions, directing shrinkage control tactics, and enhancing the durability, cost-effectiveness, and sustainability of construction processes.</p>