Prediction and Sensitivity Analysis of Reinforced Anchor Foundations Using XGBoost-SHAP Machine Learning Algorithm

Anchor foundations are frequently used for structures that experience uplift forces during their operational period. However, the prediction and analysis of the anchor uplift capacity in unreinforced and reinforced soil is challenging due to the complexities involved with the design parameters. In this paper, the application of a machine learning-based predictive model is proposed for estimating the uplift capacity of horizontally embedded anchors, both within unreinforced and geogrid-reinforced soils. The tree-based gradient boosting technique, Extreme Gradient Boosting (XGBoost), has been employed to construct the prediction model. The dataset utilised for training and evaluating the XGBoost model was obtained from a three-dimensional finite-difference numerical program. The final set of input and output datasets was fixed after examining the performance of the model by consistently varying the size of the training dataset. The Shapely Additive Explanations methodology is employed in conjunction with the XGBoost algorithm to gain a deeper understanding of the input features and their relative impact on the model's predictive outcomes. The results indicate that among the four material properties considered in the design, the anchor uplift capacity is most sensitive to the variation in soil cohesion, while the uncertainty in geogrid stiffness has the least effect. Furthermore, probabilistic analysis is also employed to incorporate the influence of uncertainties in soil and load properties on the uplift capacity and safety of the anchor. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2025.