EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

When Does Hybrid Fraud Screening Earn Its Alerts? A Budget-Constrained Comparison of Rules, Anomaly Detection, and Supervised Learning

Theo Desai

Academic Journal of Finance and Accounting · 04.11.2026 · 1 (5) · 60-72

DOI: 10.54691/r98yg421 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Academic Journal of Finance and Accounting, online erschienen am 04.11.2026. Verfasst von Theo Desai. Die Arbeit nennt 28 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Financial fraud screening operates under fixed investigation capacity, so overall discrimination may not indicate how many fraudulent transactions reach reviewers. We evaluate a two-parameter, rule-gated ranking that combines supervised histogram gradient boosting, unsupervised Isolation Forest, and three auditable red flags. Coefficients are selected exclusively on a chronological validation split to optimize fraud recall within the highest-risk 1% of transactions. Twenty-five independently seeded, explicitly synthetic streams supply 2,250,000 transactions, each with 60/20/20 temporal train/validation/test partitions. On held-out blocks, the complete hybrid averages 14.53% recall at 1% workload versus 14.08% for boosted trees (mean paired difference +0.45 percentage points; 95% seed-level t interval −0.55 to +1.45; two-sided Monte Carlo sign-flip p≈0.376). The no-rule-bonus variant still gates anomaly evidence and reaches 14.96%; a truly ungated supervised-anomaly baseline reaches 14.05%. This separates anomaly gating from direct rule-score uplift. Budget and review-window analyses show that the marginal value of explicit rule gating depends on the allocation policy and validation-selected coefficients. The central contribution is a reproducible test of when an auditable hybrid changes a restricted review queue, including cases where added components provide no benefit. Results concern simulated, not observed financial activity.

Zitieren

APA

Desai, T. (2026). When Does Hybrid Fraud Screening Earn Its Alerts? A Budget-Constrained Comparison of Rules, Anomaly Detection, and Supervised Learning. Academic Journal of Finance and Accounting, 1(5), 60-72. https://doi.org/10.54691/r98yg421

Vancouver

Desai T. When Does Hybrid Fraud Screening Earn Its Alerts? A Budget-Constrained Comparison of Rules, Anomaly Detection, and Supervised Learning. Academic Journal of Finance and Accounting. 2026;1(5):60-72. doi:10.54691/r98yg421

BibTeX
@article{Desai2026,
  author = {Desai, Theo},
  title = {When Does Hybrid Fraud Screening Earn Its Alerts? A Budget-Constrained Comparison of Rules, Anomaly Detection, and Supervised Learning},
  journal = {Academic Journal of Finance and Accounting},
  year = {2026},
  volume = {1},
  number = {5},
  pages = {60--72},
  doi = {10.54691/r98yg421}
}
Kostenlos starten →