Classification of surface settlement levels induced by TBM driving in urban areas using random forest with data-driven feature selection

Abstract

Prediction of surface settlements induced by urban area tunneling is challenging owing to the unique tunneling conditions of tunnel sites. This study presents a machine learning (ML) framework to predict the surface settlement level using a data-driven feature selection method. A large-scale database consisting of 42 settlement-influencing factors and 253 settlement measurements was acquired from a subway tunnel project in Hong Kong. The feature selection approach with three evaluation indices, i.e., predictive power score, mutual information, and feature importance, navigated the relevant features within the database. The random forest algorithm was adopted to predict the four classes of settlements defined according to their settlement levels. The efficiency of the proposed feature selection approach was verified by the accuracy and F1 score, which increased by 14.5% and 15.4%, respectively. The proposed framework can enhance the applicability of ML approaches for predicting surface settlement at complex TBM tunneling sites.