The study analyzed the MIMIC-IV database in 2,986 patients with traumatic brain injury (TBI) in the ICU, where 2,045 (68.5%) developed acute kidney injury (AKI) according to KDIGO criteria. Patients with AKI were older, heavier, had higher glucose, sodium, systolic blood pressure (SBP) and temperature, lower urine output, and more often required ventilation. Key predictors were selected using LASSO, Borut and logistic regression methods: urine output, ventilation, weight, age, glucose, sodium, SBP and temperature. Seven machine learning models were tested, with the best XGBoost achieving an AUC of 0.775 (95% CI 0.747–0.802), an accuracy of 74.4%, a sensitivity of 88.3%, and an F1-score of 0.83. Random Forest followed (AUC 0.768, sensitivity 85.9%, F1-score 0.82), while Decision Tree performed worst (AUC 0.728). SHAP analysis of XGBoost confirmed urine output and ventilation as the main predictors and provided explanations for individual patients. The XGBoost model demonstrated high discrimination, calibration and clinical benefit according to DCA, outperforming logistic regression and decision tree. It offers a tool for early AKI risk stratification in TBI patients.