With the continuous expansion of urban underground pipeline networks, ground traffic loads have imposed significant challenges to the safe operation of buried pipelines. Conventional numerical analysis methods, however, are constrained by high computational costs, making it difficult to achieve rapid and real-time pre-diction of mechanical responses. To address this gap, a sample dataset was generated through FLAC3D nu-merical simulations. On this basis, three machine learning prediction models—support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost)—were established, with their hyperparameters optimized via the Bayesian algorithm. The prediction performances of the three models for vertical displace-ments and stresses of both pipelines and surrounding soil under static and dynamic loading conditions were systematically compared. Results indicate that the prediction accuracy ranks in ascending order: SVM, RF, and XGBoost. Furthermore, quantitative analysis of feature importance was conducted, and load amplitude was identified as the dominant controlling factor governing the mechanical responses of the pipe-soil system. The study provides an efficient approach for rapid numerical estimation and parametric analysis of the mechanical responses of buried pipelines, and offers a methodological reference for the subsequent development of field measurement-based prediction models.