Forecasting Corporate Financial Distress Using Explainable Machine Learning: Implications for Risk Management Technology

Financial distress prediction is a critical task in financial risk management, yet it remains challenging due to class imbalance, complex financial structures, and limited model interpretability. This study develops an integrated machine learning framework that combines feature selection, class imbalance handling, threshold optimization, and Explainable Artificial Intelligence (XAI) to improve both predictive performance and decision-making reliability. Using a dataset of 6,819 firms, three models, Logistic Regression, Random Forest, and XGBoost, are evaluated under a consistent experimental setting. Feature selection is conducted using mutual information, while the Synthetic Minority Over-Sampling Technique (SMOTE) is applied to address class imbalance. Model performance is assessed using precision, recall, F1-score, and ROC–AUC. The results show that ensemble-based models outperform the linear baseline. XGBoost achieves the highest ROC–AUC of 0.935, while Random Forest provides more balanced performance with an F1-score of 0.421 and a recall of 0.545 for the minority class. The application of SMOTE significantly improves recall across all models, with Logistic Regression reaching a recall of 0.818. In addition, threshold optimization further enhances classification performance by improving the precision–recall trade-off. SHAP-based analysis identifies profitability and leverage-related variables as the most influential predictors, confirming consistency with financial distress theory. Overall, the findings demonstrate that integrating machine learning, handling data imbalance, and XAI leads to more reliable and interpretable financial distress prediction. The proposed framework shifts the focus from model-centric evaluation to decision-oriented financial risk assessment, providing practical value for real-world applications.