Abstract: Shanghai has entered a phase of microscale land subsidence, with its driving factors shifting from being dominated by groundwater extraction to multifactorial coupling. This study applied the XGBoost machine learning algorithm to predict subsidence in central Shanghai using monitoring data from 2020 to 2024,incorporating 12 factors across four categories: aquifer water-level variations, ground temperature changes, lithological properties, and construction activity. Factor contributions were quantified through Shapley value analysis. The results demonstrate that XGBoost achieved high prediction accuracy(R²=0.9846,RMSE=0.1240,MAE=0.0900). The XGBoost-SHAP analysis identified three primary driving factors: shallow aquifer water-level variations, temperature changes in the stable zone, and the combined effect of soft soil thickness and sandy soil ratio. Shallow aquifer water-level and stable-zone temperature changes exhibited positive correlations with settlement, while soft soil thickness and sandy soil ratio showed negative correlations. This integrated approach advances the theoretical understanding of subsidence mechanisms and supports the development of targeted prevention strategies.