<p>The quay wall, as an important component of port infrastructure, must not only bear lateral loads from vessels and seawater, but also experience lateral displacement due to soil liquefaction. This makes shear force analysis of the quay wall particularly crucial for structural design and safety assessment. However, existing methods for analyzing the shear force on quay walls perform poorly when handling large-scale datasets and nonlinear relationships. This study aims to improve the accuracy of quay wall structural shear force predictions and identify the key factors influencing shear force by combining the Light Gradient Boosting Machine (LGBM) model, shapley additive explanations (SHAP) analysis, and Sobol global sensitivity analysis. First, the LGBM model was compared with other machine learning models (XGBoost, Random Forest, and CatBoost), and the results showed that LGBM performed the best with an <i>R</i>² of 0.821, a mean absolute error of 0.065, and a root mean squared error of 0.078, demonstrating its superiority in quay wall shear force prediction. Second, SHAP analysis revealed the contribution of each input feature to the model’s predictions, and by eliminating low-impact features, the <i>R</i>² of LGBM was improved to 0.834. Finally, Sobol global sensitivity analysis further quantified the impact of each input variable on the shear force prediction, identifying the most influential features (deep soil parameters and front wall bending stiffness). This study enhances the accuracy of quay wall shear force prediction and provides scientific foundations for the safety assessment and design optimization of port infrastructure, offering significant engineering application value.</p>