The noise and nonlinear characteristics in the deformation sequence of concrete dam seriously affect the accuracy of dam deformation prediction. In this paper, ensemble empirical modal decomposition (EEMD) was used to decompose the horizontal displacement signal of the dam to mine the effective deformation information. The singular spectrum analysis (SSA) was used to extract features from the high-frequency eigenmodal components (IMF) obtained from the decomposition to reduce the loss of effective information. Considering the complex stochastic and non-linear mapping relationship between effector and environmental variables, extreme gradient boosting (XGBoost) was used to model the prediction of the noise-reduced data. Considering the significant influence of XGBoost hyperparameters on the prediction performance of the model, the Northern Goshawk algorithm (NGO) with better global search capability was introduced to perform parameter search, and an NGO-XGBoost-based dam displacement prediction model was constructed. The calculation results show that the EEMD-SSA can effectively remove the noise from the dam displacement monitoring information, and the dam deformation prediction model based on NGO-XGBoost can significantly improve the prediction accuracy.