This study aims to predict the U.S. 10-year Treasury yield using machine learning models. The selected variables include the previous day’s bond closing yields (such as the U.S. 3-month Treasury yield and the U.S. 5-year Treasury yield), U.S. financial indicators (such as the Dollar Index and the S&P 500 Index), U.S. economic indicators (such as initial jobless claims, the Core Personal Consumption Expenditures Index, and real GDP), and commodity futures (such as gold spot prices and WTI crude oil futures prices). By comparing five machine learning models—Linear Regression, Lasso, Random Forest, eXtreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR)—the results show that XGBoost performs the best in terms of Mean Absolute Error (MAE) and accuracy, followed by Random Forest. SVR demonstrates strong capability in handling high-dimensional data, while Linear Regression and Lasso models exhibit relatively weaker predictive performance. The conclusion highlights the significant advantage of the XGBoost model in predicting the U.S. 10-year Treasury yield. Future research could focus on optimizing feature selection and model parameters to enhance prediction accuracy. Additionally, incorporating more macroeconomic variables and market sentiment indicators could improve the model’s interpretability and practical value.