Abstract:

In order to improve the interpretability of excavation deformation prediction, this study developed an interpretable machine-learning model aimed at predicting the deformation of excavation retaining walls. A comprehensive analysis was conducted to evaluate the influence of different feature variables on the prediction outcomes. Firstly, a large number of excavation support structure parameters were used as a dataset, and 80% of the dataset was used to build a prediction model for the maximum lateral deflection of the retaining wall using the XGBoost (eXtreme Gradient Boosting)model. Then, the model was tested based on the remaining 20% of the dataset, and the accuracy of the model was evaluated by four indicators, i.e., the coefficient of determination, bias factor, mean absolute percentage error, and root mean square error. Finally, combined with the XGBoost model, the SHAP(SHapley Additive exPlanations) method was applied to complete the global explanation of the excavation feature variables, the partial analysis of individual samples, and the analysis of interaction effects of feature variables. The results show that the proposed method can provide both global and local explanations for the deformation prediction of excavation. At the global level, it not only provides the importance ranking of feature variables, but also gives the distribution of SHAP values. At the local level, the deformation prediction results of individual samples are decomposed into the base value and the contribution of each feature variable, which can quantify the impact of individual feature variables.

Key words: interpretability, machine learning model, excavation, retaining wall deformation, influence factor, feature variable