Resolving Interpretability Conflicts of Gradient Boosted Decision Trees (XGBoost) in Credit Risk Assessment
Journal of Organizational and End User Computing · 19.02.2026 · 38 (1) · 1-44
DOI: 10.4018/joeuc.402038 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Organizational and End User Computing, erschienen am 19.02.2026. Verfasst von Ci Zhang und 4 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 3.0).
Abstract (Originalsprache)
Credit risk assessment is a cornerstone of modern internet finance, yet the widespread adoption of machine learning models such as XGBoost is hindered by interpretability conflicts. Conventional boosting approaches often deliver strong predictive performance but produce inconsistent explanations across global and local levels, and they frequently violate domain-specific constraints such as monotonicity and fairness. To address these challenges, the authors propose UniXGB, a Unified Conflict-Resolving Explainable Gradient Boosting framework that integrates a generalized extreme value (GEV) boosting backbone with a consistency layer, multi-view explainability fusion, a conflict correction module, and an actionable counterfactual layer. The framework further supports federated adaptation and is validated through large-scale internet finance A/B testing. These findings confirm that UniXGB bridges the gap between accuracy and interpretability, offering a practical and trustworthy solution for deploying machine learning models in regulated financial environments.
Zitieren
APA
Zhang, C., Chen, H., Zhang, X., Chen, M., & Li, Y. (2026). Resolving Interpretability Conflicts of Gradient Boosted Decision Trees (XGBoost) in Credit Risk Assessment. Journal of Organizational and End User Computing, 38(1), 1-44. https://doi.org/10.4018/joeuc.402038
Vancouver
Zhang C, Chen H, Zhang X, Chen M, Li Y. Resolving Interpretability Conflicts of Gradient Boosted Decision Trees (XGBoost) in Credit Risk Assessment. Journal of Organizational and End User Computing. 2026;38(1):1-44. doi:10.4018/joeuc.402038
BibTeX
@article{Zhang2026,
author = {Zhang, Ci and Chen, Haiyu and Zhang, Xin and Chen, Min and Li, Yubo},
title = {Resolving Interpretability Conflicts of Gradient Boosted Decision Trees (XGBoost) in Credit Risk Assessment},
journal = {Journal of Organizational and End User Computing},
year = {2026},
volume = {38},
number = {1},
pages = {1--44},
doi = {10.4018/joeuc.402038}
}
EUVIMED