A hybrid SAINT–NGBoost approach for stability assessment of cantilever flood walls in anisotropic clay

: This study develops a hybrid machine-learning framework for predicting the normalized factor of safety, FS*, of cantilever flood walls embedded in anisotropic undrained shear (AUS) clay. A numerical database of 7,290 cases was generated using Finite Element Limit Analysis (FELA) in OPTUM G2, considering eight governing parameters related to soil strength, anisotropy, wall geometry, backfill configuration, and wall-soil interface conditions. The computed factor of safety was transformed into FS* , FSγ w H / (S uc0 + ρH) , representing the combined effect of hydrostatic loading and undrained soil resistance. A hybrid SAINT-NGBoost (SNGB) model was then trained and benchmarked against SVM, ANN, RF, NGBoost, and SAINT-s. The SNGB model achieved the best testing performance, with R 2 = 0.9967, RMSE = 0.2615, and MAE = 0.1437. Parametric analysis shows that L/H provides the most consistent stabilizing effect, whereas H s / H = 0.5 gives the most favorable normalized response within the investigated range. The CFI and SHAP analyses further support the dominant influence of geometric and anisotropy-related variables. A GUI was also developed for rapid forward and inverse prediction of FS* . The framework can support preliminary flood-wall stability assessment within the investigated parameter domain.