EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

FraudLens AI for Intelligent Claims Forensics and Real-Time Insurance Fraud Detection in the United States

Emmanuel Abagna

International Journal of Scientific Research in Computer Science, Engineering and Information Technology · 25.07.2024 · 10 (4) · 1161-1176

DOI: 10.32628/cseit261233040 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Scientific Research in Computer Science, Engineering and Information Technology, online erschienen am 25.07.2024. Verfasst von Emmanuel Abagna. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Insurance fraud creates substantial economic and operational pressure on the United States insurance system, while digital claims processing increases the need for fast, evidence-based screening. This study develops and evaluates FraudLens AI, a hybrid claims-forensics framework that combines supervised fraud probability, unsupervised anomaly signals, behavioral indicators, provider risk, document and narrative inconsistency, and relational-network information. The analysis uses the aggregate research dataset supplied for this study: 15,000 claims distributed across automobile/motor (40%), health/medical (30%), property/homeowners (20%), and workers' compensation (10%). The supplied class distribution contains 13,500 legitimate claims and 1,500 fraudulent or investigation-positive claims. A leakage-safe 70/15/15 training-validation-test protocol is reported, with 2,250 claims in the untouched test set. FraudLens AI is benchmarked against logistic regression, random forest, XGBoost, and an artificial neural network. On the supplied test results, FraudLens AI achieved 97.0% accuracy, 85.9% precision, 84.0% recall, an 85.0% F1-score, and a 0.967 ROC-AUC, while producing 31 false positives and 36 false negatives. Its mean inference latency was 11.6 ms per claim, corresponding to approximately 86 claims per second. The strongest normalized features were claim amount/severity (0.16), anomaly score (0.14), narrative inconsistency (0.12), and provider risk (0.11). These findings indicate that multi-signal fusion can improve fraud prioritization relative to the benchmark models in the supplied results. Because the source material provides aggregate outcomes rather than the underlying 15,000 row-level records, the reported model results are analyzed as supplied and are not presented as an independently reproduced training experiment.

Zitieren

APA

Abagna, E. (2024). FraudLens AI for Intelligent Claims Forensics and Real-Time Insurance Fraud Detection in the United States. International Journal of Scientific Research in Computer Science, Engineering and Information Technology, 10(4), 1161-1176. https://doi.org/10.32628/cseit261233040

Vancouver

Abagna E. FraudLens AI for Intelligent Claims Forensics and Real-Time Insurance Fraud Detection in the United States. International Journal of Scientific Research in Computer Science, Engineering and Information Technology. 2024;10(4):1161-1176. doi:10.32628/cseit261233040

BibTeX
@article{Abagna2024,
  author = {Abagna, Emmanuel},
  title = {FraudLens AI for Intelligent Claims Forensics and Real-Time Insurance Fraud Detection in the United States},
  journal = {International Journal of Scientific Research in Computer Science, Engineering and Information Technology},
  year = {2024},
  volume = {10},
  number = {4},
  pages = {1161--1176},
  doi = {10.32628/cseit261233040}
}
Kostenlos starten →