To improve the prediction accuracy of muck volume in slurry shield tunneling, field data from the Suzhou Yangcheng West Lake Third Channel Branch Tunnel Project were used. A muck volume estimation method based on storage tank liquid level variations was proposed by considering the relationships among storage tank liquid levels, make up water, separated muck, and filter pressed waste slurry. Its rationality was verified through an analysis of construction parameter variations. Cutterhead rotational speed, advance rate, inlet and outlet slurry densities and flow rates, slurry chamber pressure, and air cushion chamber pressure were selected as input variables. A time lag-genetic algorithm-extreme gradient boosting (TL-GA-XGBoost) model was developed for muck volume prediction, and Shapley additive explanations (SHAP) were applied to analyze the effects of the input variables. The results indicate that the proposed estimation method can reflect the operating condition of the slurry circulation system. Higher prediction accuracy was achieved by the TL-GA-XGBoost model than by the comparison models. Based on the SHAP results and parameter variation patterns during stable tunneling, proposed ranges for the main construction parameters were identified. The findings can provide a reference for the safety of slurry shield tunneling.