Prediction of Ground Subsidence Risk in Urban Centers Using Underground Characteristics Information

Abstract

Ground subsidence primarily occurs due to complex factors, such as damage to underground facilities and excavation work, and its occurrence can result in loss of life and damage to property. Therefore, factors that induce ground subsidence must be investigated to prevent accidents. This study aims to evaluate and predict the ground subsidence risk in urban centers in South Korea. To this end, a machine learning-based ground subsidence risk prediction model was constructed by utilizing data on the underground facility attribute information, permeability coefficient, stratigraphic thickness, and height. The random forest, XGBoost, and LightGBM machine learning algorithms were used to develop the prediction model, and the SMOTE sampling technique was employed to address data imbalance. The reliability of the developed model was verified using the evaluation metrics of F1-score and accuracy. The best-performing model was selected to create a risk map and visualize the areas with ground subsidence risk. The results indicate that the incorporation of additional data improves model performance and reliability. Thus, the machine learning model with various factors developed in this study offers foundational insights for the prevention and risk management of ground subsidence.