Drift-Aware Explainable Supplier Risk Prediction: Monotonic LightGBM, Probability Calibration, and Cost-Sensitive Auditing
International Journal of Business and Social Science Studies · 28.10.2026 · 6 (2) · 331-340
DOI: 10.66795/ijbsss63 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Business and Social Science Studies, online erschienen am 28.10.2026. Verfasst von Lena Anderson und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Supplier screening is operationally consequential only when its predictions remain defensible and its rankings allocate scarce review capacity effectively. Building on constrained, explainable supplier-risk modeling, we develop a drift-aware protocol that combines monotone LightGBM, TreeSHAP attribution, validation-only probability calibration and severity-weighted audit ranking, and jointly tests directional coherence, temporal probability shift and loss capture under limited audit budgets. On a fully specified cohort of 10,000 supplier records with 18 predictors, five ordered periods and ex-ante event severity, monotone constraints eliminate all 462 counterfactual direction violations of unconstrained LightGBM across 4,900 probes while raising ROC-AUC from 0.8230 to 0.8275, and Platt calibration lowers the Brier score from 0.1701 to 0.1652 on the final 2,000-record test period. Under a 10% audit budget, severity-weighted ranking captures 52.08% of realized loss versus 21.23% for probability-only ranking, and regularized logistic regression provides a strong, transparent reference for both discrimination (0.8401) and expected-loss ranking. The protocol offers a reproducible, decision-centered basis for validating supplier-risk models on institutional data.
Zitieren
APA
Anderson, L., & Jain, O. (2026). Drift-Aware Explainable Supplier Risk Prediction: Monotonic LightGBM, Probability Calibration, and Cost-Sensitive Auditing. International Journal of Business and Social Science Studies, 6(2), 331-340. https://doi.org/10.66795/ijbsss63
Vancouver
Anderson L, Jain O. Drift-Aware Explainable Supplier Risk Prediction: Monotonic LightGBM, Probability Calibration, and Cost-Sensitive Auditing. International Journal of Business and Social Science Studies. 2026;6(2):331-340. doi:10.66795/ijbsss63
BibTeX
@article{Anderson2026,
author = {Anderson, Lena and Jain, Omar},
title = {Drift-Aware Explainable Supplier Risk Prediction: Monotonic LightGBM, Probability Calibration, and Cost-Sensitive Auditing},
journal = {International Journal of Business and Social Science Studies},
year = {2026},
volume = {6},
number = {2},
pages = {331--340},
doi = {10.66795/ijbsss63}
}
EUVIMED