Hybrid Financial Representation Stacking Ensemble For Bankruptcy Prediction

This study presents a Hybrid Financial Representation Stacking Ensemble for explainable corporate bankruptcy prediction. The framework integrates original financial variables with traditional financial ratios and structural composition features, then adds an Autoencoder anomaly score, a Temporal 1D-CNN, CatBoost, an XGBoost stacking meta-learner and SHAP explanations. The approach is designed to address the main difficulties identified in the project: complex financial relationships, temporal deterioration, highly imbalanced bankruptcy data, redundant variables and the need for interpretable predictions. Using 78,682 firm-year observations from the 1999–2018 period, the proposed HFRSE framework achieved 72.12% Accuracy, 11.88% Precision, 55.29% Recall, 19.56% F1-score and 0.7015 ROC-AUC. The Temporal 1D-CNN achieved the highest Recall, while CatBoost achieved the highest Accuracy, Precision, F1-score and ROC-AUC. The stacking framework combined these strengths and produced a balanced result. SHAP further showed that Market Value, Retained Earnings, Net Income to Revenue, Debt Ratio, ROA and the Autoencoder Anomaly Score are among the most influential factors. The results suggest that HFRSE can serve as an early-warning and financial-risk assessment framework, particularly where decision-makers need both a prediction and an explanation. Future extensions can incorporate additional economic and business characteristics, richer disclosure or textual information, improved temporal architectures, calibrated probabilities and cost-sensitive approaches to further improve minority-class precision and practical deployment.

Ataş, P. K., Tak, N., Özöğür-Akyüz, S., & Erdogan, B. E. (2025). An innovative approach to ensemble learning in bankruptcy prediction using support vector machines and meta fuzzy functions. Information Sciences, 719, Article 122450. Ansari, A., Ahmad, I. S., Bakar, A. A., & Yaakub, M. R. (2020). A hybrid metaheuristic method in training artificial neural network for bankruptcy prediction. IEEE Access, 8, 176640–176655. Chen, X., Liu, J., & Wu, C. (2025). Multi-class financial distress prediction based on hybrid feature selection and improved stacking ensemble model. Expert Systems With Applications, 282, Article 127832. Gabrielli, G., Melioli, A., & Bertini, F. (2026). Corporate financial distress prediction: A machine learning approach in the era of big data. Journal of Accounting & Organizational Change, 22(7), 31–65. Gaurav, A., Gupta, B. B., Bansal, S., & Psannis, K. E. (2025). Bankruptcy forecasting in enterprises and its security using hybrid deep learning models. Cyber Security and Applications, 3, Article 100070. Hussain, A. B., Jingchun, S., Hussain, B., Shaique, M., & Rehman, A. U. (2026). Cross-market bankruptcy prediction: An interpretable ensemble learning framework using SHAP analysis. Computational Economics. Kadkhoda, S. T., & Amiri, B. (2024). A hybrid network analysis and machine learning model for enhanced financial distress prediction. IEEE Access, 12, 52759–52777. Lakshmi, T. M., Martin, A., & Venkatesan, V. P. (2016). A genetic bankrupt ratio analysis tool using a genetic algorithm to identify influencing financial ratios. IEEE Transactions on Evolutionary Computation, 20(1), 38–51. Lombardo, G., Bertogalli, A., Consoli, S., & Reforgiato Recupero, D. (2024). Natural language processing and deep learning for bankruptcy prediction: An end-to-end architecture. IEEE Access, 12, 151075–151091.