EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Enhancing Credit Risk Assessment in Consumer Lending: An E‐MLP Approach With Uncertainty Quantification

Cuiling Zhu, Yunpeng Wang, Tantan Liang, Chengwei Yang, Mingyuan Hu

International Journal of Intelligent Systems · 2026-01 · 2026 (1)

DOI: 10.1155/int/3595080 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Intelligent Systems, online erschienen am 24.07.2026. Verfasst von Cuiling Zhu und 4 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Key Technology Research and Development Program of Shandong, Humanities and Social Science Fund of Ministry of Education of China, Social Science Planning Project of Shandong Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper proposes a credit risk assessment method based on a deep integrated multilayer perceptron (E‐MLP) framework. This method not only retains the core advantages of MLP, such as robust nonlinear fitting ability and efficient feature learning effect, but also integrates a three‐module architecture composed of estimator, calibrator, and discriminator. Firstly, the estimator will generate the initial point estimation of the probability of default through multiple different MLP submodels, so as to improve the stability of the prediction. Second, based on the Bayesian principle, the calibrator uses second‐order moment correction to refine the uncertainty estimation and then solves the inherent limitations of traditional MLPs in point prediction. Third, the discriminator aggregates the submodel output through a weighted integration strategy to capture data heterogeneity while optimizing overall accuracy. The empirical results on the LendingClub dataset show that this method achieves significant performance improvement. Compared with logistic regression (LR), random forest (RF), and extreme gradient boosting (XGB), the E‐MLP model improves the area under the curve (AUC) by about 1.19%, 3.38%, and 0.34%, respectively. In addition, the model also has excellent calibration performance. After the application of sigmoid, beta, and isotonic calibrators, the average accuracy, and specificity are improved by 0.595% and 0.315%, respectively, compared with the uncalibrated baseline. On the Prosper dataset, the E‐MLP model is also superior to LR, RF, CatBoost, and ordinary MLP. The AUC is increased by 4.79%, 4.83%, 0.15%, and 0.86%, respectively, and the accuracy after calibration is also increased by 1.801% on average. By providing accurate point estimates and reliable uncertainty intervals, the E‐MLP framework creates a deployable and interpretable solution that fills the gap between advanced deep learning techniques and actual credit risk management needs.

Förderung

  • Key Technology Research and Development Program of Shandong
  • Humanities and Social Science Fund of Ministry of Education of China
  • Social Science Planning Project of Shandong Province

Zitieren

APA

Zhu, C., Wang, Y., Liang, T., Yang, C., & Hu, M. (2026). Enhancing Credit Risk Assessment in Consumer Lending: An E‐MLP Approach With Uncertainty Quantification. International Journal of Intelligent Systems, 2026(1). https://doi.org/10.1155/int/3595080

Vancouver

Zhu C, Wang Y, Liang T, Yang C, Hu M. Enhancing Credit Risk Assessment in Consumer Lending: An E‐MLP Approach With Uncertainty Quantification. International Journal of Intelligent Systems. 2026;2026(1). doi:10.1155/int/3595080

BibTeX
@article{Zhu2026,
  author = {Zhu, Cuiling and Wang, Yunpeng and Liang, Tantan and Yang, Chengwei and Hu, Mingyuan},
  title = {Enhancing Credit Risk Assessment in Consumer Lending: An E‐MLP Approach With Uncertainty Quantification},
  journal = {International Journal of Intelligent Systems},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/int/3595080}
}
Kostenlos starten →