Abstract: The fatigue failure of stainless steel round square mixed tubes may occur under cyclic load. A " prediction model of stress concentration factor" with high prediction accuracy and explanation mechanism was proposed to quantitatively evaluate the fatigue resistance of stainless steel round square mixed tube joints. A database of 487 stainless steel tube joints (T-, Y-, X- and K-joints) was created and key input characteristics were identified. Then, nine typical regression algorithms were selected to establish the model, and the predicted values of each model were compared with the test values, so as to obtain the optimal prediction model of stress concentration factor. On this basis, Shapley method was used to explain the global, individual and feature dependencies. The results showed that the prediction accuracies of both test set and training set of XGBoost model were greater than 0.98, and the feature selection was effective. The XGBoost model had high prediction accuracy and generalization ability when predicting the SCF of stainless steel round square mixed tube joints. The influence of overlap and overlap rate should be considered in SCF prediction formula and there was a high dependence between the dimensional features of the section. The human-computer interactive GUI module can accurately predict the stress concentration coefficient of stainless steel round square mixed tube joints.