Gradient boosting decision tree (GBDT) for firm failure prediction is proposed. Sensitivity analysis and model interpretability of GBDT are analyzed and validated. GDBT, bagging, Adaboost, Random ...