The authors systematically review the progress of machine learning applications in predicting surface settlement caused by shield tunneling, focusing on input parameters, prediction objectives, algorithms selection, and hyperparameter optimization. Key challenges are identified, and future research directions are proposed. The findings include the following: (1) Integration of tunnel geometric parameters, stratum properties, and shield machine operation parameters constitutes the predominant research focus for settlement prediction. (2) Selecting suitable models and input parameters tailored to specific prediction objectives is critical. (3) Intelligent hyperparameter optimization can significantly enhance prediction accuracy. However, current studies face several limitations: (1) Most models lack the ability to autonomously identify relevant features and are susceptible to overfitting; (2) Analysis and utilization of large-scale datasets remain inadequate; (3) Robust models leveraging multi-source heterogeneous datasets are yet to be developed; and (4) Research on predicting the developmental processes of surface settlement is relatively scarce. Finally, critical issues requiring attention in advancing intelligent shield tunneling are discussed.