Abstract: To address the issues of structural damage to existing tunnels and operational safety risks of high-speed railways caused by excessive deformation of high-speed railway tunnels due to shield tunneling underneath, accurate prediction of the maximum deformation of overlying tunnels induced by shield tunnel excavation is crucial. Based on a multi-source heterogeneous database of disturbance deformation from shield tunneling underneath existing tunnels, this study constructed a general prediction model for deformation of overlying high-speed railway tunnels induced by new shield tunneling underneath using machine learning algorithms including XGBoost, support vector machine (SVM), and neural network.Taking the project case as the research object, the XGBoost algorithm was used to model and predict the settlement of high-speed railway tunnels induced by shield tunneling underneath. Bayesian algorithm was employed to optimize the model hyperparameters, and the recommended range of tunneling parameters for shield tunneling underneath was proposed based on the settlement limits of high-speed railway tunnels.The results show that: (1) The XGBoost model optimized by Bayesian algorithm was used to establish a surrogate model for predicting disturbance of existing tunnels induced by shield tunneling. Over 95% of the prediction errors of this model are less than 0.65 mm, with the coefficient of determination (R²) reaching 0.997, which is significantly better than the SVM and neural network models. (2) By combining the prediction surrogate model with the particle swarm optimization (PSO) algorithm, an optimization control model with decision variables including shield tunneling speed, cutter head rotation speed, synchronous grouting pressure, and synchronous grouting volume was established. The optimization of tunneling parameters was implemented for the shield tunneling section underneath the Liuyang River Tunnel to guide construction, and the maximum settlement of the ballastless track structure was 1.12 mm, meeting the track deformation control standards. (3) The temporal variation law of the predicted settlement values of the high-speed railway tunnel after optimization of shield tunneling parameters is consistent with that of on-site monitoring values, with a maximum deviation of 0.198 mm, verifying the practicability and accuracy of the disturbance prediction and tunneling parameter optimization control method for shield tunneling underneath existing tunnels. The research results can further guide the determination of the range of tunneling parameters for shield tunneling underneath high-speed railway tunnels and provide a theoretical reference for ensuring the safe operation of high-speed railways.