Evaluating feature contribution to wall deflection of braced excavation using interpretable XGBoost-SHAP model

It is of great significance to interpret machine learning algorithms in predicting excavation deformation. This paper develops an interpretable machine learning model for evaluating feature contribution to wall deflection induced by braced excavation. The impact of various features on the model predictions has been analyzed in detail. Initially, a dataset consisting of a large number of maximum wall deflections is utilized, with 80% of the dataset employed to construct a prediction model for the maximum wall deflection using the eXtreme Gradient Boosting (XGBoost) algorithm. Subsequently, the model is tested using the remaining 20% of the dataset, and its accuracy is evaluated using four performance indicators. Finally, in combination with the XGBoost model, the SHapley Additive exPlanations (SHAP) method is applied to evaluate the relative importance of the features associated with maximum wall deflection, and reveal how each feature affects the model prediction. This study can provide important insights and guidance for identifying key factors in the deformation control of excavation supporting structures.