Financial Risk Warning and Traceability for Manufacturing Enterprises: A Three-Stage Framework Based on Ensemble Learning Algorithm

Manufacturing enterprises operate in an increasingly volatile global supply chain environment, making financial risk early warning a critical prerequisite for ensuring sustainable operations. Current research primarily focuses on qualitative categorizations of financial distress, neglecting systematic indicator processing, and quantitative risk prediction. Moreover, the lack of interpretability in complex prediction models limits their practical applicability. Therefore, we propose a three-stage framework based on ensemble learning that includes risk indicator preprocessing, financial risk prediction and interpretability analysis, aiming at early warning and traceability of financial risks. In the first stage, we introduce a compensated entropy weight method to calculate the financial risk value (FRV), incorporating layered indicator screening for preprocessing. This method ensures robust weight assignments even under entropy extremes, enhancing the quantitative assessment of financial risk. In the second stage, an improved whale optimization algorithm (IWOA)-enhanced extreme gradient boosting (XGBoost) model is proposed for financial risk prediction. This approach overcomes the tendency to converge to local optima, achieving superior predictive accuracy with a mean absolute percentage error (MAPE) of 13.55%, outperforming existing models. The third stage employs fuzzy C-means clustering (FCM) to classify risk levels based on FRV, followed by Shapley additive explanations (SHAP) analysis to identify key influencing factors and their evolving patterns across different risk levels. These three stages support each other and progress step by step to realize the targeted financial risk control of manufacturing enterprises.