Abstract: Addressing the limitations of traditional algorithms for predicting the friction resistance coefficient (α) in mine tunnels—such as low prediction accuracy, susceptibility to local optima during optimization, and slow convergence—this paper proposes an Improved Bald Eagle Search algorithm (IBES) optimized Extreme Gradient Boosting (XGBoost) prediction model, termed IBES-XGBoost. The IBES algorithm enhances the standard Bald Eagle Search (BES) by integrating Opposition-Based Learning (OBL) to improve initial population quality, introducing a chaotic sequence-driven adaptive parameter adjustment strategy to balance global exploration and local exploitation, implementing a dynamic adaptive mutation strategy to maintain population diversity and avoid premature convergence, and incorporating Chaotic Local Search (CLS) to refine the search for optimal solutions. Testing on standard benchmark functions confirms that IBES significantly outperforms the original BES and other metaheuristic algorithms in solution accuracy, convergence speed, and stability. Leveraging this efficient IBES, the key hyperparameters of the XGBoost model are adaptively optimized, utilizing multi-dimensional tunnel characteristics as inputs and minimizing the Root Mean Square Error (RMSE) as the objective function. Experimental results demonstrate that the IBES-XGBoost model excels in friction resistance coefficient prediction, achieving a test set RMSE of 0.001232, MAE of 0.000868, and an R2 of 0.985426. This model significantly outperforms all comparison models; specifically, compared to the suboptimal BES-XGBoost model, it reduces RMSE and MAE by 49.94% and 49.09% respectively, demonstrating superior prediction accuracy and robustness.