The prediction of shield tunneling attitude holds significant importance for shield construction. Existing research primarily focuses on the application of deep learning in time-series prediction, while the practical value of traditional machine learning methods and confidence interval prediction in engineering applications is often overlooked. In addition, the determination of the number of modes in modal decomposition remains uncertain, which limits model stability and prediction accuracy. To achieve better prediction of shield attitude, based on the shield construction project of the Jiangyin-Jingjiang Yangtze River Tunnel, SSA (sparrow search algorithm) was applied to jointly optimize the number of modes and iterations of VMD (variational mode decomposition), enabling adaptive and efficient decomposition of non-stationary signals. Furthermore, SSA was utilized to optimize the hyperparameters of SVR (support vector regression), constructing a high-precision time-series prediction model. The results indicate that the test set performance of the SSA-VMD-SVR model achieves MSE=0.484 36, RMSE=0.680 78, and R²=0.969 58, outperforming traditional models such as VMD-BILSTM and EMD-SVR. Moreover, prediction uncertainty is quantified through confidence interval analysis based on kernel density estimation, shifting from “predictable results” to “quantifiable risks”. This study provides a high-precision prediction tool and scientific decision-making support for shield tunnel construction under complex geological conditions.