Application of tree-based methods in predicting the surface settlement arising from the tunnel excavation with large mix-shield

Surface settlement due to tunnel excavation is affected by several factors. However, no explicit mapping correlation exists between surface settlement and the main influencing factors. In this study, three tree-based methodologies, including classification and regression tree (CART), random forest (RF), and gradient boosting decision tree (GBRT), were implemented to predict the tunneling-induced surface settlement of the South Hong-Mei Road tunnel in Shanghai, where a large mix-shield was used. Thirteen influencing factors within three categories (tunnel geometry, geological conditions, and shield operation factors) were employed as input variables. Results show that the ensemble methods (RF and GBDT) provide superior performance over the single-tree model (CART). Moreover, GBDT has the highest level of prediction accuracy among the three statistical learning methods. The importance of influencing factors on the tunneling-induced surface settlement was probed. The tunnel geometry had the greatest effect on surface settlement. It is followed by the influencing factors in shield operation factors. Moreover, geological conditions were not as influential as the other influencing factors. The outcomes of this study may provide a reference for evaluating tunneling-induced surface settlement in other similar tunnel projects.