Predicting Financial Distress via Static and Dynamic Features: A Boruta-Enhanced XGBoost Approach with SHAP Interpretability

Early identification of corporate financial distress is essential to reduce systemic risk and maintain market stability. However, most existing studies rely on financial indicators at a single point in time, making it difficult to capture changes in firms’ operating conditions and complex nonlinear patterns. As a result, risk signals are often detected too late, and model outcomes are hard to interpret. This study proposes an integrated prediction framework that combines static financial indicators with dynamic growth indicators to construct a time-sensitive feature system. Using data from Chinese A-share listed firms between 2007 and 2024, the Boruta algorithm is first applied to identify key predictors and reduce noise from high-dimensional data. An XGBoost model is then developed, with hyperparameters automatically optimized through Optuna to better handle class imbalance and nonlinear relationships. To enhance model transparency, SHAP is employed to interpret prediction results and assess their financial consistency. Empirical results show that the proposed model significantly outperforms traditional linear models and commonly used ensemble methods. The optimized XGBoost model achieves an accuracy of 0.939, an AUC of 0.964, and a PR-AUC of 0.870, demonstrating strong capability in identifying rare financial distress cases. SHAP analysis indicates that dynamic indicators, such as year-over-year net profit growth, often play a more important role than traditional measures (e.g., Altman’s Z-score). In addition, the model effectively detects profit–cash flow mismatches, where rapid profit growth unsupported by operating cash flow signals a higher risk of financial distress. Corporate financial distress is rarely a sudden event; instead, it usually results from the long-term accumulation of valuation misalignment, declining earnings quality, and weakening growth momentum. By integrating indicators that reflect both structural stability and dynamic trends, this framework provides a more forward-looking prediction tool. It not only improves predictive accuracy but also offers actionable insights for credit rating and investment decisions by distinguishing sustainable growth from superficial expansion. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2026.