DEVELOPING EXPLAINABLE AI SCORECARDS FOR RETAIL CREDIT RISK: A REPRODUCIBLE FRAMEWORK USING LENDING DATA
DOI:
https://doi.org/10.65164/q81mq324Ключевые слова:
credit scoring, scorecard, synthetic data, weight of evidence & information values, regularized logistic regression, explainable AI, machine learning, NPL, ARАннотация
This article presents a step‑by‑step framework for developing an interpretable
scorecard for retail credit‑risk assessment in banks and financial institutions, and validates the
approach on a fully synthetic consumer‑lending dataset. The model is an integration of
Weight‑of‑Evidence & Information Value (WoE & IV) binning, machine learning algorithm – Lasso
(L1) Regularized Logistic Regression, model‑risk supervision, statistical metrics such as Roc-auc &
Gini and Confusion Matrix. The testing dataset demonstrates ROC‑AUC of 86% and a 72% Gini
uplift generalization over a rule‑based benchmark. The final result is a ready credit scorecard
illustrating the distribution of points assigned to the background features of a customer applying for
a loan.
Библиографические ссылки
[1] Lin, A. Z., & Stokes, M. (2015). Entropy-based measures of weight of evidence and
information value for variable reduction and segmentation for continuous dependent
variables. SAS Global Forum 2015 Proceedings, Paper 3242-2015.
https://support.sas.com/resources/papers/proceedings15/3242-2015.pdf
[2] Lin, A. Z. (2014). Expanding the use of weight of evidence and information value.
Southeast SAS Users Group (SESUG) 2014 Proceedings, Paper SD-20.
https://www.lexjansen.com/sesug/2014/SD-20.pdf
1310
[3] Bhalla, D. (2015, March). Weight of evidence (WOE) and information value (IV) explained.
Listen Data. https://www.listendata.com/2015/03/weight-of-evidence-woe-andinformation.
html
[4] Anik, C. (2020). Weight of evidence (WoE) and information value (IV): How to use it in
EDA and model building. Medium. https://anikch.medium.com/weight-of-evidence-woe-andinformation-
value-iv-how-to-use-it-in-eda- and model-building-3b3b98efe0e8
[5] Analytics Vidhya. (2021, June).Understand weight of evidence and information value.
https://www.analyticsvidhya.com/blog/2021/06/understand-weight-of-evidence-andinformation-
value/ SAS Institute Inc. (n.d.).
[6] Computing weight of evidence and information value in PROC CASBINNING. SAS
Visual Statistics Documentation.
https://documentation.sas.com/doc/en/vdmmlcdc/8.1/casstat/viyastat_binning_details02.htm
urafsky, D., & Martin, J. H. (2025).
[7] Speech and language processing: An introduction to natural language processing,
computational linguistics, and speech recognition (3rd ed.).
https://web.stanford.edu/~jurafsky/slp3/
[8] Divya, D., Shrivarshini, J., Karan, S., & Kanimozhi, R. (2024). Artificial intelligence in
finance. ResearchGate.
https://www.researchgate.net/publication/380266484_ARTIFICIAL_INTELLIGENCE_IN_FI
NAN CE
[9] Kelly, B. T., & Xiu, D. (2023). Financial machine learning (Working Paper No. 2023-100).
Becker Friedman Institute, University of Chicago. https://bfi.uchicago.edu/wpcontent/
uploads/2023/07/BFI_WP_2023-100.pdf
[10] Frontiers of Business Research in China. (2020). Toward understanding weight of
evidence and information value. FBR, 14(1), Article 6.
https://fbr.springeropen.com/articles/10.1186/s11782-020-00082-6
