Abstract: The potential risk evaluation of the reservoir dam infrastructures is an important part of the reservoir dam risk evaluation system.However, the monitoring data of reservoir dam infrastructures is large with many characteristics and easy to lose.As an integrated algorithm based on decision tree, XGBoost algorithm has unique advantages in dealing with feature data with large-scale missing data and mixed type.Therefore, in order to evaluate the potential risk of reservoir dam infrastructures quickly and accurately, this paper proposed a potential risk assessment and prediction method for reservoir dam infrastructures based on XGBoost.Firstly, the reservoir dam monitoring data was preprocessed, and the XGBoost model was trained with the data.Then, the optimal parameters of the model were calculated by GridSearch and Cross-validation.Finally, the model was evaluated according to accuracy indicators such as accuracy and recall rate.The prediction results showed that the accuracy of XGBoost on the test set reached 91.26%,which was 2.12%,5.59%,19.31% and 38.65% higher than the other four conventional machine models(random forest, artificial neural network, nearest neighbor algorithm and support vector machine).The proposed model can meet the requirements of engineering practice.