Time-series interferometric synthetic aperture radar （TS-InSAR） technology has been widely used in ground deformation monitoring and prediction. However，current research remains insufficient in the correlation and temporal lag between groundwater and ground deformation. Moreover，InSAR-based prediction models for ground deformation mostly rely on a single InSAR data，which limits the prediction accuracy and generalization ability of the models. To address these challenges，this study proposed a combination-long short-term memory （C-LSTM） model that integrates groundwater level，rainfall，and InSAR deformation data. This model was employed to evaluate the prediction and accuracy of single-factor and multi-factor models，respectively. The results revealed a temporal lag between ground deformation and changes in groundwater level. The optimal feature combination，obtained through model training using groundwater and rainfall data，exhibited significant improvements in prediction accuracy compared to single-factor predictions，with the coefficient of determination （R²） increasing by 2.45%，1.52%，4.16%，8.08%，5.08%，and 1.45% respectively. The model enhances the prediction accuracy of ground deformation by incorporating model feature combinations with high correlation with ground deformation.