Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM12164
Volume:
6 |
4 |
Apr 2022
Pages:
1-12
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Database Developer
Group Disability and Life Insurance domain
Background. Insurance fraud leaks roughly $308 billion a year out of U.S. carriers, and the life and life-and-health line is the largest single slice — about $74.7 billion. Coordinated fraud rings, where policyholders, beneficiaries, agents, and medical examiners collude across multiple policies, drive a disproportionate share of that loss. The peer-reviewed Scopus literature on graph-based fraud-ring detection has been built almost entirely on auto, credit-card, and social-security data; life-line work is empty. Production claims systems run on relational substrates such as IBM DB2 and Oracle, and the academic in-database graph-analytics thread has been quiet since MADlib in 2012. Methods. We present a portable PL/SQL recursive-CTE traversal pattern that materialises a six-entity life-insurance graph (Policy, Policyholder, Beneficiary, Agent, Medical Examiner, Death Claim) with seven node-level ring-signal features: beneficiary-policy multiplicity, agent early-death rate, examiner death-rate z-score, policy-to-death cycle days, beneficiary–agent co-occurrence count, two-hop component size, and approximate betweenness centrality. The schema runs on IBM DB2 LUW and Oracle 19c with cross-substrate output equivalence to six decimal places. We benchmark four classical ring-detection methods over the materialised graph: Louvain community detection, k-core decomposition (k≥4), a FRAUDAR-style dense-subgraph detector, and connected-component anomaly scoring. Threshold calibration uses ExCeeD per-example confidence applied at the ring level. The pipeline runs end-to-end on a synthetic life-insurance graph of 50,000 policies, 30,000 beneficiaries, 5,000 agents, 2,000 examiners, and 750 planted fraud rings. Results. The schema is method-stable. Cross-method top-50 ring overlap by Jaccard runs from 0.62 to 0.85 across the six method pairs, with Louvain–FRAUDAR pairing the highest at 0.85 and connected-component-anomaly–Louvain the lowest at 0.62. The same five top-attribution features (beneficiary-policy multiplicity, agent early-death rate, beneficiary–agent co-occurrence, policy-to-death cycle days, two-hop component size) carry dominant signal across all four methods. Mean recall at the top 50 ring candidates is 0.755 averaged across methods, with FRAUDAR-style detection highest at 0.84. ExCeeD-calibrated thresholds cut false-positive operational cost by 35.5% on average compared to top-N quantile thresholds at matched recall. DB2 and Oracle produce equivalent feature values; recursive-CTE traversal scales near-quadratically in edge count, with 10k-, 100k-, and 1M-edge runtimes of 4.8 s, 71.2 s, and 21.4 minutes respectively. Conclusions. The substrate-resident graph schema, not the detection method, is the load-bearing reproducibility artifact for life-insurance fraud-ring detection. The schema is portable across DB2 and Oracle, requires no external graph-database infrastructure, and gives any future life-line ring-detection paper a citable baseline to extend. The reference implementation is released as supplementary material.
KEYWORDS
This paper proposes a portable PL/SQL recursive-CTE schema for life-insurance fraud-ring detection, validated against four classical ring-detection methods. The empirical finding is that the schema, not the detection method, is the load-bearing reproducibility artifact: cross-method top-50 Jaccard is at least 0.62 across all pairs, per-feature attribution is stable across method families, and ExCeeD-calibrated thresholds reduce false-positive operational cost by 35.5% on average versus top-N quantile thresholds at matched recall. The view layer materialises with cross-substrate output equivalence between DB2 and Oracle to six decimal places, and recursive-CTE traversal scales near-quadratically to one million edges. The contribution is durable. Graph-detection algorithms evolve; the schema interface is stable. Practitioners on the operational side of group life insurance can adopt the view layer with low integration cost and obtain a defensible operational baseline. The fact that no peer-reviewed Q1–Q2 paper has previously published a life-insurance ring-detection schema is, itself, the strongest motivation for the contribution. Several extensions are natural. Real-data validation with carrier partners would promote the artifact to Gold reproducibility tier. Temporal extensions via time-weighted networks per GOTCHA [14] would catch rings that evolve across the contestability period rather than appearing as static structure. Cross-line generalisation to property and workers' compensation would test whether the six-entity schema ports cleanly into other lines. GNN-based extensions per [11, 15] are a candidate paper-#3 or paper-#4 angle if carrier data becomes available. Privacy-preserving cross-carrier ring detection — federated graph analytics, differential privacy on shared edges — would unlock the long-standing problem that ring fraud routinely spans multiple carriers and no single carrier has the full graph. The reference implementation is released as supplementary material under MIT license. Practitioners on the operational side of group life insurance are invited to fork the schema, run it against their own warehouses, and report findings.
[1] L. Akoglu, H. Tong, and D. Koutra, "Graph based anomaly detection and description: A survey," Data Mining and Knowledge Discovery, vol. 29, no. 3, pp. 626–688, 2015. DOI: 10.1007/s10618-014-0365-y. [2] T. Pourhabibi, K.-L. Ong, B. H. Kam, and Y. L. Boo, "Fraud detection: A systematic literature review of graph-based anomaly detection approaches," Decision Support Systems, vol. 133, p. 113303, 2020. DOI: 10.1016/j.dss.2020.113303. [3] B. Hooi, H. A. Song, A. Beutel, N. Shah, K. Shin, and C. Faloutsos, "FRAUDAR: Bounding graph fraud in the face of camouflage," in Proc. ACM KDD 2016, pp. 895–904, 2016. DOI: 10.1145/2939672.2939747. [4] V. D. Blondel, J.-L. Guillaume, R. Lambiotte, and E. Lefebvre, "Fast unfolding of communities in large networks," Journal of Statistical Mechanics: Theory and Experiment, vol. 2008, no. 10, p. P10008, 2008. DOI: 10.1088/1742-5468/2008/10/P10008. [5] M. E. J. Newman, "Modularity and community structure in networks," PNAS, vol. 103, no. 23, pp. 8577–8582, 2006. DOI: 10.1073/pnas.0601602103. [6] S. Fortunato, "Community detection in graphs," Physics Reports, vol. 486, no. 3-5, pp. 75–174, 2010. DOI: 10.1016/j.physrep.2009.11.002. [7] L. Šubelj, S. Furlan, and M. Bajec, "An expert system for detecting automobile insurance fraud using social network analysis," Expert Systems with Applications, vol. 38, no. 1, pp. 1039–1052, 2011. DOI: 10.1016/j.eswa.2010.07.143. [8] M. Óskarsdóttir, W. Ahmed, K. Antonio, B. Baesens, R. Dendievel, T. Donas, and T. Reynkens, "Social Network Analytics for Supervised Fraud Detection in Insurance," Risk Analysis, vol. 42, no. 8, pp. 1872–1890, 2022. DOI: 10.1111/risa.13693. [9] M. Tumminello, A. Consiglio, P. Vassallo, R. Cesari, and F. Farabullini, "Insurance fraud detection: A statistically validated network approach," Journal of Risk and Insurance, vol. 90, no. 2, pp. 381–419, 2023. DOI: 10.1111/jori.12415. [10] I. Vorobyev, "Fraud risk assessment in car insurance using claims graph features in machine learning," Expert Systems with Applications, vol. 251, p. 124109, 2024. DOI: 10.1016/j.eswa.2024.124109.
Publication Date
4/26/2022
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
0
Total views since publication
PDF Downloads
1
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...