This study aims to predict concrete total shrinkage in Thailand and analyze its characteristics. It consists of three key objectives. Firstly, it assesses the suitability of various machine learning algorithms and finds that the Random Forest Regressor (RF) and Extreme Gradient Boosting (XGB) are the most compatible models for all time-dependent shrinkage test data in Thailand. Secondly, the Leave-One-Out prediction method enhances the study’s robustness from small datasets of total shrinkage at specific ages compared to the current empirical prediction model. Feature importance analysis identifies critical parameters, including cement content, water content, superplasticizer amount, and slump. Notably, machine learning-derived insights indicate that only some key parameters can yield predictions with a less than 10% error, facilitating practical and efficient prediction of total shrinkage. Lastly, the study analyzes the significance of fly ash properties, including chemical components and found %LOI consistently being influential. Despite limited data, reducing parameters based on importance scores proves effective, suggesting benefits in computational efficiency and model interpretability. Overall, this research offers valuable insights for improving concrete total shrinkage prediction in Thailand.