Dams play a pivotal part in national economic development, and to protect the operation of dams, monitoring and prediction of dam deformation are crucial. In response to the fact that the commonly used HST (hydrostatic, season, time) model for predicting dam deformation in the past cannot explain the influence of environmental variables well, the measured temperature was used to replace the harmonic function in the HST model as the temperature component, and the lagged water levels were added as an influencing factor. Then, the feature set was filtered using the Recursive Feature Elimination (RFE) algorithm, which reduces the dimensionality of the original feature set to the same level as the HST model to alleviate the computational burden of the model and the Support Vector Machine (SVM) was used to build the model. Aiming at the sensitivity of SVM model prediction accuracy to parameters, the Particle Swarm Optimization algorithm was introduced to find the best combination of parameters for SVM, and the SVM-RFE dam deformation prediction model was established. The measured data of Shuibuya concrete-faced rockfill dam verified the validity of the model. The final analysis indicate that compared with the traditional model, the improved SVM-RFE dam deformation prediction model obviously improves the prediction performance, reflecting the superiority of the model.