A machine learning framework based on improved extreme gradient boosting for predicting material removal depth

Material Removal Depth (MRD) is a key indicator that determines the surface quality and machining accuracy in robotic grinding processes. To address the issues of insufficient accuracy and the “black-box” nature of models in current MRD prediction methods, this study proposes an MRD prediction approach that integrates an Improved Egret Swarm Optimization Algorithm (IESOA) and eXtreme Gradient Boosting (XGBoost). The method enhances prediction performance by optimizing the hyperparameters of XGBoost using IESOA and improves model interpretability by incorporating Shapley Additive exPlanations (SHAP). Experimental results demonstrate that the proposed IESOA-XGBoost model achieves high predictive accuracy. Compared to XGBoost, SVR, RF, LightGBM, IESOA-SVR, IESOA-RF, and IESOA-LightGBM, its prediction accuracy is improved by 7.08%, 9.92%, 18.88%, 7.59%, 8.75%, 15.52%, and 4.36%, respectively. SHAP analysis further quantifies the influence of each process parameter on MRD, providing a clear theoretical basis for optimizing grinding process parameters and enhancing the credibility of model decision-making.