The splitting tensile strength of concrete plays a crucial role in maintaining structural integrity, as tensile stresses induced by loading and environmental variations commonly lead to cracking. This study investigates the effectiveness of advanced ensemble Machine Learning (ML) models, including LightGBM, Gradient Boosting Regression Tree (GBRT), Extreme Gradient Boosting (XGBoost), and AdaBoost, in accurately predicting the splitting tensile strength of silica-fume-modified concrete. Using a comprehensive database divided into training (80%) and testing (20%) sets, the performance of the models was evaluated using the R², Root Mean Square Error (RMSE), and MAE metrics. The results indicate that the GBRT and XGBoost models achieved superior predictive accuracy, with R² scores reaching 0.999 during the training stage and high accuracy during the testing stage (XGBoost: R² = 0.965, RMSE = 0.337; GBRT: R² = 0.955, RMSE = 0.381), outperforming the LightGBM and AdaBoost models. This study introduces GBRT and XGBoost as reliable and efficient alternatives to conventional experimental methods, offering significant savings in time and cost. In addition, SHapley Additive exPlanations (SHAP) analysis was performed to identify the key input features and clarify their effects on splitting tensile strength, providing valuable insights into the predictive behavior of silica-fume-modified concrete. The SHAP analysis revealed that the water-to-binder ratio and curing duration were the most influential factors affecting the splitting tensile strength of Silica-Fume (SF) concrete.