ABSTRACT
During earth pressure balanced (EPB) shield tunnel boring machine (TBM) excavations, accurately predicting over-excavation is crucial for implementing timely mitigation measures. However, previous studies have faced challenges in reliably predicting scenarios with a higher risk of over-excavation. This study proposes a data-driven approach to enhance over-excavation detection using machine learning techniques combined with data augmentation. The synthetic minority oversampling technique (SMOTE) was employed to address the class imbalance between normal and over-excavation instances. A random forest (RF) model was subsequently developed to classify these instances. The RF model with SMOTE exhibited superior performance in detecting over-excavation instances, whereas the RF model without SMOTE failed to identify even half of these instances. These comparative results demonstrate the importance of data augmentation in addressing class imbalance and improving the detection of over-excavation.
