Hybrid ensemble models for enhanced slope stability prediction based on a comprehensive database
•The largest open-source slope stability database was created, containing 897 cases.•Two hybrid ensemble frameworks achieved 92.22 % accuracy, outperforming existing models.•The study explores the models’ applicability across different slope types.•An open-source GUI was developed for easy deploymen...
Uloženo v:
| Vydáno v: | Transportation Geotechnics Ročník 56; s. 101816 |
|---|---|
| Hlavní autoři: | , , , , |
| Médium: | Journal Article |
| Jazyk: | angličtina |
| Vydáno: |
Elsevier Ltd
01.01.2026
|
| Témata: | |
| ISSN: | 2214-3912, 2214-3912 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | •The largest open-source slope stability database was created, containing 897 cases.•Two hybrid ensemble frameworks achieved 92.22 % accuracy, outperforming existing models.•The study explores the models’ applicability across different slope types.•An open-source GUI was developed for easy deployment in engineering projects.•Four outlier handling methods and eight hybrid models were evaluated to select the optimal base model.
Traditional methods for slope stability assessment often encounter limitations in prediction accuracy and adaptability when applied to complex geological conditions and nonlinear characteristics typical of mining areas. To address these challenges and enhance analytical efficiency, this study develops a comprehensive slope stability database comprising 897 cases. Based on this dataset, we proposed two intelligent prediction frameworks grounded in hybrid ensemble learning utilizing the hiking optimization algorithm to perform hyperparameter optimization for both light gradients boosting machine and extreme gradient boosting models. Considering the randomness of the optimization algorithm, a new base model pairing strategy is proposed. For each base learner, we constructed three optimized configurations resulting in six base models. The models are then integrated using stacking and voting strategies to form hybrid stacking and hybrid voting models. All models are evaluated using five-fold cross-validation. The hybrid ensemble models achieve an accuracy of 0.9222, both significantly outperforming base models. Moreover, this study also explores the optimal base model selection strategy and compares the applicability of the models across different slope types. The interpretability analysis using SHapley Additive exPlanations (SHAP) reveals that the internal friction angle (φ) and cohesion (C) are the most influential factors governing stability prediction across both hybrid frameworks. To support practical engineering applications, we developed a user-friendly graphical user interface, enhancing the operability and applicability of the models. Comparative analysis demonstrates that the proposed method significantly improves the accuracy of slope stability prediction, outperforming traditional machine learning approaches. Overall, this research achieves key advancements in both database construction and intelligent prediction methodology, offering theoretical foundations and technical pathways for slope stability assessment under complex geological conditions. |
|---|---|
| ISSN: | 2214-3912 2214-3912 |
| DOI: | 10.1016/j.trgeo.2025.101816 |