Abstract: A machine learning study on the hydrodynamic characteristic parameters of a tandem three-cylinder flow at low Reynolds numbers was carried out based on the (XGBoost) eXtreme Gradient Boosting algorithm and SHAP (SHapley Additive exPlanations) analysis. The open-source computational fluid dynamics software OpenFOAM was employed to simulate and establish datasets of lift and drag forces and vortex shedding frequencies of each cylinder under various conditions. By comparing parameters such as the coefficient of determination, absolute error, and error rate, the machine learning model based on the XGBoost algorithm, after hyper-parameter optimization, exhibited excellent predictive performance. In the prediction of literature parameters outside the dataset range, the maximum error rate was 16.03%, which could be reduced to 0.71% after secondary learning. By using SHAP analysis to explain the model's prediction results both globally and locally, the cumulative average contribution degrees of Reynolds number, upstream spacing, and downstream spacing to the nine hydrodynamic characteristic parameters of the tandem three-cylinder configuration were obtained, and an attribution analysis was conducted. Additionally, the local contribution value variations of the input features were captured. Through combined analysis with the flow field structure, it was discovered that when the upstream spacing was 2 and the downstream spacing increased from 2 to 3, the SHAP value of the downstream spacing to the average drag force of the downstream cylinder increased from −0.22 to 0.03, and the SHAP value to the root mean square lift force increased from −0.22 to 0.04. It was discovered that the wake interference pattern transformed from an expanded body to an alternating reattachment pattern. When the upstream spacing was 6 and the downstream spacing increased from 2 to 6, the SHAP local analysis quantified the variation laws of the hydrodynamic characteristics of the downstream cylinder in the double-row vortex structure.