Dam failure often causes an enormous loss of life and property and huge environmental damage. Accurate and fast estimation of the risk level of earth-rock dams is of great significance for controlling their failure hazards. This paper develops a fast prediction model of the earth-rock dam risk grade based on GA-LightGBM

using the K-Nearest Neighbor (KNN) algorithm to fill a large amount of missing data in the database

and adopting a Genetic Algorithm (GA) to optimize the hyperparameters of Light Gradient Boosting Machine (LightGBM). The model accuracy is verified using the receiver operating characteristic (ROC) curves

the area under the curve (AUC)

and other evaluation indexes; and it is compared with the traditional machine learning model. The results show that this new model has a high accuracy of 89.95% and its AUC value is 0.977

indicating it is better in terms of applicability and accuracy. Analysis of global influencing factors and case studies using Shapley Additive Explanations (SHAP) show the frequency of inspection is one of the most important factors leading to the risk of earth-rock dams.