Due to the continuous rise of urban underground space development, there has been an increasing number of deep excavation project in recent years, in China. The complexity of hydrogeological conditions and surrounding environment makes it difficult to calculate the deformation of enclosure structure caused by deep excavation. Based on the monitoring data during the entire deep excavation process, four data-driven machine-learning algorithms were developed and evaluated in this study to predict deep horizontal displacement (DHD) of enclosure structure considering the excavation process and spatial effect. The results showed that the XGBoost algorithm has the best performance, with an RMSE of 2.3397, an MAE of 1.5732 and R2 of 0.9088. Furthermore, it has higher accuracy in predicting the DHD at the corners, followed by the deformation caused by excavation stage 2. It can be seen that the machine-learning algorithms can be a potential tool for predicting DHD caused by deep excavation in practical engineering.