A tool wear prediction method based on the improved sparrow search algorithm (ISSA) optimized XGBoost model is proposed to enhance the tool wear prediction performance under different machining conditions. The recursive feature elimination (RFE) algorithm is applied to adaptively identify the most representative features from cutting forces and torque signals. The XGBoost model is utilized to establish the tool wear prediction model, with ISSA optimizing its hyperparameters. The experimental results demonstrate that the RFE algorithm reduces the feature dimension by 94.22% and improves the model’s precision by 10.64%. The proposed ISSA-XGBoost model has high recognition accuracy, with a classification accuracy of 93.62%. Comparative analyses with prediction models optimized using particle swarm optimization (PSO), grey wolf optimization (GWO), and sparrow search algorithm (SSA), the ISSA-XGBoost model achieves the improvements of 4.26%, 10.64%, and 8.51%, respectively, allowing it to efficiently identify the tool wear state at various spindle speeds.