Reconstruction of the administrative boundaries of Yuxian during the Ming Dynasty based on machine learning technology

The Ming Dynasty Great Wall functioned as an ideal military boundary, showing strong spatial coupling among military settlements, pass systems, and natural geography. To address the subjectivity of traditional weight-based methods (e.g., PCA, AHP), this study applies machine learning, combined with textual analysis and spatial modeling, to reconstruct Yuxian’s Ming-era administrative boundaries. Results show that the XGBoost model surpasses logistic regression and random forest in prediction accuracy, enhancing objectivity and reliability. SHAP explanations reveal nonlinear interactions among cost factors, with dominant influences including high elevation, proximity to beacon towers, and distance from main roads. The reconstructed boundaries align well with historical descriptions of mountains, valleys, and passes, confirming traditional logic of “defense on high ground, avoiding rivers, and delimiting by landmarks.” This study presents a replicable approach to historical geographic reconstruction and provides data support for cultural heritage conservation and spatial governance.