Sepsis is an important cause of death in hospitals, especially for patients in intensive care unit (ICU). Early identification of individuals with a high risk of poor prognosis in sepsis patients is of great significance, and timely and appropriate treatment can markedly enhance patient prognosis. A retrospective study design was adopted in this paper. First, sepsis patients who meet the definition of sepsis 3.0 were selected from the MIMIC-IV database, and combined with voting method, prediction models with different advantages were screened out. Then, the voting with stacking STE was constructed to predict the 30-day mortality risk of sepsis patients. Finally, the effectiveness of the proposed model was tested and contrasted with traditional models. Besides, the prediction results of the STE-model was explained by the SHAP method. The accuracy, precision, recall, specificity, F1 score and AUC of the proposed STE-model are 0.890, 0.870, 0.890, 0.977, 0.870 and 0.889, respectively, and the performance of STE-model is better than other methods. The comprehensive results demonstrate STE-model in promptly identifying patients at heightened risk of adverse outcomes, and can provide practical, assistance for clinicians to take accurate management and treatment programs.