The study built and validated machine learning models to predict in-hospital mortality in patients with sepsis-associated acute kidney injury (S-AKI). They analyzed data from the MIMIC-IV 3.0 database in adult ICU patients according to Sepsa-3 (suspected infection with SOFA score ≥2) and KDIGO criteria, supplemented by a prospective cohort from Ningxia Hospital in 2023–2025. Predictors included demographics, comorbidities, vital signs, laboratory results, and severity scores within 24 hours of ICU admission. The XGBoost model achieved the best performance with an AUC of 0.8799 in internal validation, with high sensitivity, precision and F1 scores. SHAP analysis identified SAPS II score, AKI stage, oxygenation index, serum sodium, and blood urea nitrogen as major risk factors. External validation confirmed the robustness of the XGBoost model. The model enables early identification of high-risk patients and personalized management in sepsis care.