<p>Durability is one of the most important criteria for a concrete mix. Water absorption of hardened concrete is a property that directly impacts the service life of concrete products directly, although it is a simple concept to understand, it is difficult to compare the results of different experiments and reach a conclusion. This study aimed to develop models using machine learning approaches to predict water absorption of hardened concrete immersed in water. Thorough research was conducted on reputable published literature. After extensive standardization, a total of 266 usable and validated data rows were selected for use in developing the models. Models were developed based on the EXtreme Gradient Boosting (XGBoost) algorithm and were optimized using Grid Search, Random Search, and Bayesian Optimization methods. The performance and accuracy of each model were assessed by using different criteria. Although all three models performed admirably, the XGB-Grid Search model performed best on the training set of data with an R² of 0.99997, and closely trailing behind were XGB-Random Search (R² = 0.99959) and XGB-Bayesian Optimization (R² = 0.99946). On the testing dataset, however, the XGB-Random Search model achieved the best performance with an R² of 0.99710, followed by XGB–Bayesian Optimization (R² = 0.99557) and XGB-Grid Search (R² = 0.99436). Considering all the facts and analysis of the three developed models, the XGB-Random Search model proved to be the overall best performer, successfully achieving the study’s goal of predicting the water absorption of concrete. The impact of the input variables was analyzed using SHapley Additive exPlanations (SHAP) analysis to determine their influence on the model development.</p>