Abstract: The demand for waterway freight volume is influenced by numerous factors. Following the implementation of the "645" project in the midstream section of the Yangtze River, the navigation conditions have significantly improved. To better analyze the trend changes in freight volume after the project implementation, this study introduces a novel model for forecasting waterway freight volume. Initially, the quadratic interpolation method and the KNN inverse distance weighting interpolation method are employed to address issues of inconsistency in time granularity and missing data in high-dimensional panel data. By utilizing hierarchical clustering and the interpretability of SHAP values, key influence factor feature sequences are comprehensively screened to reduce the dimensions and scale of input data for the forecasting model. The introduction of the Halton low-discrepancy sequences and the Quasi-Reflective Bayesian Learning (QRBL) strategy substantially enhances the optimization efficiency of the Tuna Swarm Optimization (TSO) algorithm, improving the TSO algorithm's optimization effectiveness of hyperparameter combinations, such as the number of decision trees, the depth of decision trees, and learning rate in the eXtreme Gradient Boosting (XGBoost) model that determine the model's fitting ability. The results indicate that the new model significantly outperforms comparative models in forecasting accuracy, demonstrating better applicability for waterway freight volume forecasting under the influence of multiple feature factors.