Noto Peninsula Earthquake on 1 January 2024, caused liquefaction damage in various parts of Niigata Prefecture. Particularly, in Niigata City’s Nishi Ward, road surface depressions and sand boil were observed. Given this fact, Kochi Prefecture conducted evacuation drills assuming road liquefaction in preparation for the anticipated Nankai Trough Megathrust Earthquake. The results highlighted issues such as the slow evacuation speed of the elderly and wheelchair users and the disruption of evacuation routes due to ground subsidence. This underscores the importance of wide-area predictions of ground subsidence caused by liquefaction for evacuation planning. In this context, our research focuses on digital national ground information and Japan Engineering Geomorphologic Classification Map (JEGM) as well as the N-value obtained from boring surveys to show the trend patterns of ground subsidence and make a broader prediction. Here, we aim to develop a ground subsidence prediction model for the purpose of improving liquefaction hazard maps during the major earthquakes throughout Japan. We use a machine learning algorithm based on XGBoost (eXtreme Gradient Boosting), constructing a single prediction model by sequentially combining weak learners based on Gradient Boosting Decision Tree. Additionally, the impact of each explanatory variable on the prediction value assessed using SHAP (SHapley Additive exPlanations) for an engineering explanation of the model. This enables high-speed computation even with large datasets and complex expressions, considering interaction variables. According to the results, the features that had a significant impact on ground subsidence were seismic intensity, maximum elevation, and AVS30. Since SHAP values indicate the contribution to the target variable, it was found that these variables contribute up to 0.06 to 0.09 m to ground subsidence.