This dataset is derived from the master's thesis, “Predictive Analysis of Pipe Surge Risks in Yangtze River Binary Embankment Based on Gray Correlation and GA-DBN”. In that study, the deep belief network (GA-DBN) method optimized by genetic algorithm was used to apply and validate the levee leakage prediction model by importing five levee leakage factors of Yangxin Dry Dike of the Yangtze River, which were collected as practical application cases, into the model for analysis.

  (1) A Deep Belief Network (DBN) model and a Genetic Algorithm optimized DBN (GA-DBN) model are used for the prediction of embankment leakage, which can effectively capture the complex features in the data through the multilayer nonlinear structure. The DBN model can effectively capture the complex features in the data, while the GA-DBN model further combines the genetic algorithm to optimize the network structure and parameters, which improves the prediction accuracy and generalization ability of the model.

  (2) The data are input into the support vector machine (SVM) model and compared and analyzed with the training results of the GA-DBN model.

  The data quality evaluation based on the training results of the GA-DBN model indicates that the present dataset is of extremely high quality and reliability. During the validation process, the GA-DBN model achieved 100% accuracy, recall, precision, and F1 score, which further validated the high quality of the dataset and the excellent predictive ability of the model, which was significantly better than the SVM model. These excellent performance indicators show that the integrity and accuracy of the data are fully guaranteed during the selection of influence factors, data collection and processing, and can effectively support the prediction and analysis of embankment seepage risk.