The study developed an interpretable machine learning (Gradient Boosting Machine) model to predict spontaneous bleeding in children with pediatric acute liver failure (PALF). The model was trained on the data of 501 patients from the Children's Hospital of Chongqing Medical University and externally validated on 153 patients. It achieved an AUC of 0.858 in internal validation (95% CI, 0.778–0.899) and an AUC of 0.839 in external validation (95% CI, 0.774–0.904). Key predictors included platelet count, infection, multiorgan dysfunction syndrome (MODS), hepatorenal syndrome (HRS), D-dimer, total protein, and lactic acid. The SHAP analysis showed that infection, MODS and HRS increase the risk of bleeding, while higher platelet count, protein and fibrinogen are protective. The model demonstrated good calibration and clinical utility according to the decision curve.