This study investigates the role of polyethylene glycol (PEG) as an internal curing agent in concrete using data mining techniques. A total of 1200 data points were collected from reputable journal publications, and after preprocessing steps such as filtering, normalization, and outlier removal, the dataset was refined to 723 reliable records. Five machine-learning models—Linear Regression (LRM), Decision Tree, Random Forest, XGBoost, and Support Vector Machine (SVM)—were trained to predict the compressive strength of PEG-modified concrete. Model performance was assessed using MAE, RMSE, R², and SI. Among the tested algorithms, XGBoost demonstrated the highest predictive accuracy, achieving an MAE of 0.496 and an RMSE of 0.626, making it both the most reliable and computationally efficient model for forecasting the compressive strength of internally cured concrete. Feature-importance analysis further revealed that PEG’s molecular weight plays a more significant role than its dosage. Higher-molecular-weight PEGs (e.g., PEG 4000 and PEG 6000) delivered greater improvements in strength, even at low replacement levels, compared to lower-molecular-weight variants. In high-strength concrete with a water-to-cement ratio below 0.33, controlling PEG content is crucial to ensure sufficient internal curing water. The combined use of PEG and silica fume—up to 8% replacement—substantially enhanced the density of the C–S–H gel, resulting in notable gains in both compressive strength and durability. Overall, this study presents an XGBoost-based predictive framework for evaluating PEG-induced internal curing effects. The findings offer practical guidance for designing efficient PEG-based internal curing systems, reducing water demand, and supporting concrete production in water-limited regions.