<![CDATA[Predicting the ultimate bearing capacity (qult) of strip footings on slopes remains a major challenge in geotechnical engineering, as classical methods were developed for level ground and lose reliability for complex slope-foundation geometries. Although finite element analysis (FEA) provides better accuracy, its high computational cost limits large-scale parametric studies. This study proposes a hybrid FEA–machine learning (ML) framework to estimate qult of strip footings on slopes, overcoming the accuracy limitations of existing analytical solutions for different slope-foundation configurations. The dataset of 600 finite element simulations was developed under the Mohr-Coulomb plane strain constitutive framework. Six variables were examined: unit weight (γ), cohesion (c), friction angle (φ), applied load (P), foundation width (B), and embedment depth (Df). Seven predictive models were developed: multiple linear regression, polynomial regression, support vector regression, decision trees, random forests, k-nearest neighbors, and extreme gradient boosting (XGBoost). Model performance was assessed using R², RMSE, MAPE, and the a20 index, with R² and RMSE as the primary ranking criteria, while Shapley Additive Explanations (SHAP) were applied to interpret feature contributions. XGBoost has the highest prediction accuracy on both the training and test datasets. It is followed by Support Vector Regression (SVR). The most influencing parameter in all seven models was the foundation depth (Df), followed by the friction angle (φ) and the foundation width (B), while the slope angle consistently decreased the predicted bearing capacity. The results confirm the accuracy, interpretability, and computational efficiency of the integrated FEA-ML approach as an alternative to traditional bearing capacity analysis.]]>
Copyrights © 2026
