Monitoring and Predicting Deformation in Deep Excavation: A Comparative Study of Supervised Learning Methods

Abstract Accurate deformation prediction is critical for managing risks in deep excavation projects, particularly under real‐world conditions where data quality issues like noise, data loss, and irregular time intervals are prevalent. This study addresses the gap in existing research by evaluating six supervised learning methods—SVM, LSTM, GRU, Elman, BP, and a convolutional neural network with BP—through three robustness experiments designed to simulate real‐world challenges. The convolutional neural network with BP demonstrated superior performance, achieving the lowest MAE and RMSE, particularly in noise and data loss scenarios. These findings highlight its potential to significantly enhance safety in excavation monitoring. Further refinement and development could broaden its application to a wider range of geotechnical engineering problems.