为解决地铁盾构隧道施工引起的地表沉降预测过程中数据样本不足、对数据预处理较为粗糙的问题，选取北京地区32个地铁盾构隧道施工引起的地面沉降数据作为数据库，采用合成少数类过采样技术(synthetic minority oversampling technique, SMOTE)算法对数据库进行扩增，并在此基础上选取BP神经网络(back propagation, BP)、随机森林(random forest, RF)、支持向量机(support vector machine, SVM)和K近邻(K-nearest neighbor, KNN)4种机器学习模型对沉降进行预测分析。分析结果表明： 1）经过预处理后的数据集预测能力显著增强，其中，KNN模型表现最佳，测试集平均绝对误差仅为1.60 mm； 2）采用KNN模型对北京轨道交通12号线西坝河—三元桥区间地层沉降进行预测，基于该模型预测值的Peck公式与实测值拟合度较高； 3）基于数据增强下的KNN模型对于盾构施工引起的地面沉降变形有良好的预测效果。

The prediction of surface settlement induced by shield tunneling has many problems such as insufficient data samples and relatively rough data preprocessing. As a result, a case study is conducted on Beijing metro shield tunneling projects, and the surface settlement induced by shield tunneling of 32 cases are selected. Then, the synthetic minority oversampling technology is used to enlarge the database, and four machine learning models, namely back propagation, random forest, support vector machine, and knearest neighbor (KNN), are employed to predict the surface settlement. The comparative analysis shows the following: (1) The prediction ability of the preprocessed database is significantly enhanced, and KNN shows a smallest average absolute error of the test set, which is 1.60 mm. (2) The bestperforming model is applied to XibaheSanyuanqiao shield tunneling section of Beijing metro line 12, and the predicted results agree well with the monitored results. (3) The KNN model based on the data enhancement has a good surface settlement prediction effect.