Reliability-Gated Temporal Heterogeneous Graph Learning for Credit Default Prediction
International Journal of Business and Social Science Studies · 06.10.2026 · 6 (2) · 299-309
DOI: 10.66795/ijbsss81 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Business and Social Science Studies, online erschienen am 06.10.2026. Verfasst von Wei Anderson und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Credit default prediction benefits from temporal financial histories and context-dependent relations. Building on the temporal–structural decomposition of DeepCreditRisk, we develop a reliability-gated temporal heterogeneous learning architecture that combines chronological GRU encoding with training-only borrower–state prototypes. Relation groups are assigned from raw-scale delinquency and utilization measurements, a featurewise reliability gate conditions each relation message on the borrower’s temporal state, and prototypes are estimated from training borrowers only, so that no held-out outcome enters the relational context. Experiments on three independently generated populations of 12,000 borrowers with borrower-disjoint splits evaluate AUROC, AUPRC, Brier score, gate ablations and temporal-order interventions. The gated model attains a mean AUROC of 0.82952 ± 0.00281, on par with logistic regression and the temporal GRU, and reversing the monthly order lowers its AUROC from 0.82776 to 0.63676 (seed 42), confirming that the architecture exploits chronological evidence. The resulting leakage-aware typed-relation framework provides a transparent basis for credit models that incorporate observed financial-event networks.
Zitieren
APA
Anderson, W., Cohen, S., & Chandra, K. (2026). Reliability-Gated Temporal Heterogeneous Graph Learning for Credit Default Prediction. International Journal of Business and Social Science Studies, 6(2), 299-309. https://doi.org/10.66795/ijbsss81
Vancouver
Anderson W, Cohen S, Chandra K. Reliability-Gated Temporal Heterogeneous Graph Learning for Credit Default Prediction. International Journal of Business and Social Science Studies. 2026;6(2):299-309. doi:10.66795/ijbsss81
BibTeX
@article{Anderson2026,
author = {Anderson, Wei and Cohen, Sabrina and Chandra, Kevin},
title = {Reliability-Gated Temporal Heterogeneous Graph Learning for Credit Default Prediction},
journal = {International Journal of Business and Social Science Studies},
year = {2026},
volume = {6},
number = {2},
pages = {299--309},
doi = {10.66795/ijbsss81}
}
EUVIMED