Suicide is a major public health concern in the United States, with county-level variations influenced by socioeconomic, demographic, health, and environmental factors. Data indicate that about 22% of U.S. counties saw a rise in suicide rates between 2016 and 2023, with approximately 8% of counties reporting an increase of 50% or more. Accurate suicide prediction models can help identify high-risk regions and support targeted suicide prevention, intervention, and control efforts. This study aims to contribute to this effort by benchmarking eleven machine learning models for predicting county-level suicide rates across the U.S., utilizing publicly available datasets covering the eight-year period from 2016 to 2023. The evaluated models include tree-based methods (XGBoost, Random Forest, CatBoost, LightGBM, Gradient Boosting, and Decision Tree), distance-based learning (K-Nearest Neighbors), regression models (Linear Regression, Ridge Regression, and Lasso Regression), and Support Vector Regression. Results indicate that XGBoost achieved the highest predictive accuracy (R2 = 0.97) with relatively low RMSE and MAE, followed by Random Forest and CatBoost, highlighting the effectiveness of these tree-based ensemble learning techniques. In contrast, traditional regression and distance-based learning models underperformed, demonstrating limitations in capturing the complexity of suicide prediction. This benchmarking study underscores the importance of advanced ensemble methods in predictive modeling for public health applications. The findings provide a foundation for future research and policy-driven initiatives aimed at suicide prevention, emphasizing the need for data-driven decision-making at the county level.