Structure Learning-Based Interaction Feature Construction for Bank Failure Prediction
IEEE Access · 2026 · 14 · 75131-75144
DOI: 10.1109/access.2026.3689132 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Chiwoo Lee und 2 weiteren Autor:innen. Die Arbeit nennt 23 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Research Foundation of Korea (NRF) grants funded by the Korea government.
Abstract (Originalsprache)
Machine-learning models for bank failure prediction commonly rely on bank-level financial ratios. Although such models often achieve strong predictive performance, they offer limited insight into the joint signaling of distress by multiple indicators. This research introduces a dependency-guided interaction feature framework for bank failure prediction. The framework employs structure learning to identify dependency relationships among financial ratios and to transform them into edge-level interaction features. These features are then combined with the original ratios and incorporated into standard predictive models, without presuming causal validity. Empirical evaluation on a real-world bank failure dataset with severe class imbalance demonstrates that dependency-guided interaction features consistently enhance predictive discrimination and robustness when combined with original financial indicators. Furthermore, the explainability analysis indicates that the proposed representation facilitates relationship-level explanations, complementing traditional feature-level interpretations. This study contributes to establishing structure learning as a principled and model-agnostic feature-construction mechanism that enhances both predictive performance and interpretability in bank failure prediction.
Förderung
- National Research Foundation of Korea (NRF) grants funded by the Korea government
Zitieren
APA
Lee, C., Pyo, S., & Cho, M. (2026). Structure Learning-Based Interaction Feature Construction for Bank Failure Prediction. IEEE Access, 14, 75131-75144. https://doi.org/10.1109/access.2026.3689132
Vancouver
Lee C, Pyo S, Cho M. Structure Learning-Based Interaction Feature Construction for Bank Failure Prediction. IEEE Access. 2026;14:75131-75144. doi:10.1109/access.2026.3689132
BibTeX
@article{Lee2026,
author = {Lee, Chiwoo and Pyo, Sujin and Cho, Minsu},
title = {Structure Learning-Based Interaction Feature Construction for Bank Failure Prediction},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {75131--75144},
doi = {10.1109/access.2026.3689132}
}
EUVIMED