A heterogeneous ensemble framework, termed the slope-stability hybrid ensemble network (SHEN), was proposed for high-precision prediction of the factor of safety (FOS) in slope stability. The framework was engineered to model the complex nonlinear physical relationships inherent in slope stability analysis. Within this framework, Bayesian optimization was first applied to tune the hyperparameters of extreme gradient boosting (XGBoost) and light gradient boosting machine (LightGBM), followed by weighted integration based on their performance on the validation set. Subsequently, a secondary dynamic fusion was performed using a deep neural network (DNN). Based on a database of 302 excavation soil slope engineering cases (including characteristics such as heavy γ, cohesion c, internal friction angle ϕ, slope angle β, slope height H, pore pressure ratio ru, etc.), the regression and classification of the model were evaluated on a 10 fold cross validation and independent test set, with FOS (factor of safety) ≥ 1.0 as the stability performance. The assessment was carried out through 10-fold cross-validation and an independent test set. The results demonstrate that superior performance was demonstrated by SHEN on the test set. R²of 0.907 0, RMSE of 0.148 9, and MAE of 0.107 4 are achieved for regression, an F₁ of 0.937 5 is obtained for classification. All individual benchmark models are outperformed. Individual XGBoost, LightGBM, and DNN models are outperformed by SHEN. The feature selection capabilities of tree-based models are integrated with the nonlinear representation abilities of neural networks in this approach. A reusable technical framework for stability assessment and risk management in geotechnical engineering is provided by this method.