An innovative UBFEM–ANN approach for slope stability evaluation and slip surface prediction of frictional soil slopes

To address the limitations of existing machine learning approaches in simultaneously predicting slope stability and failure characteristics, this study proposes an efficient framework for slope failure characteristic prediction by integrating upper-bound finite element limit analysis (UBFEM) with an artificial neural network (ANN). A local six-node, sixfold-path discontinuous triangular mesh is employed to exploit the complementary advantages of plastic elements and velocity discontinuities in critical slip surface identification. The sliding body is subsequently identified based on velocity discontinuities, enabling the construction of a high-fidelity dataset of primary slope failure characteristics. A dataset comprising 1008 parameter combinations of classical homogeneous frictional soil slopes is used to train the machine learning models. The results demonstrate that the optimized model maintains excellent predictive performance across different slope geometries and soil strength conditions, achieving an 𝑅 2 of 0.999 and an RMSE of 0.356 for critical unit weight prediction, while the mean relative error MRE of the sliding body area prediction is only 0.036. Furthermore, the Hausdorff distance is employed to quantify error propagation during the reconstruction of the critical slip surface, while permutation feature importance (PFI) and SHapley additive explanations (SHAP) are combined to quantitatively evaluate the effects of input parameters on slope stability and slip surface geometry. The proposed UBFEM–ANN framework therefore provides an effective and interpretable approach for predicting slope stability, critical slip surface geometry, and sliding body area.