EIGF: An Explainable Graph Neural Network Fusion Approach for Fraud Risk Detection in Insurance Claims

Tiejiang Sun; Xu Han; Jiaying Chen; Yao Ge

doi:10.6911/

Authors

Tiejiang Sun Chang’an University, Xi’an, China
Xu Han California State Polytechnic University, Pomona, United States
Jiaying Chen Cornell University, Ithaca, United States
Yao Ge University of Miami, Coral Gables, United States

DOI:

https://doi.org/10.6911/

Keywords:

Insurance fraud, Medicare claims, graph neural networks, GraphSAGE, explainable artificial intelligence, class imbalance, evidence fusion.

Abstract

Screening insurance claims for fraud poses three difficulties at once. Positives are scarce, the entities under review are linked to one another, and every alert consumes investigator time, so a workable system has to rank well and show its evidence. We describe EIGF, an Explainable Insurance Graph Fusion framework in which a tabular risk prior is corrected by relational signal from a GraphSAGE branch. The data are public Medicare provider records released under CC BY 4.0. We collapse 558,211 inpatient and outpatient claims into 60 leakage-controlled provider features covering 5,410 providers, 506 of whom carry a potential-fraud label. Shared beneficiaries and shared physicians give rise to two provider graphs, each built with inverse-entity-degree weighting followed by top-k pruning. Class-weighted focal loss trains the graph branch; a balanced histogram gradient-boosting branch captures nonlinear tabular risk. A convex weight, tuned on validation precision-recall area under the curve (PR-AUC) alone, mixes the two probabilities and lets the model back away from noisy graph propagation when that helps. Over five stratified 60/20/20 splits EIGF reaches a test PR-AUC of 0.7211±0.0358, ROC-AUC of 0.9464±0.0114, F1 of 0.6554±0.0287, and recall of 0.7663±0.0429. Mean PR-AUC gains are 0.0102 over histogram gradient boosting and 0.0103 over GraphSAGE. Against the graph branch the gain holds on every split (one-sided exact Wilcoxon p=0.03125); against the tabular branch the direction is consistent, but five splits cannot settle it. Our explanation package draws on SHAP, integrated gradients, relation ablation, neighborhood evidence, and a masking-based fidelity test. Zeroing the five highest-ranked features cuts the mean risk probability of high-confidence fraudulent providers by 0.3839, against 0.0549±0.0727 for random five-feature masks. Two further findings matter for practice: beneficiary-sharing edges carry more information than physician-sharing edges, and on some splits the edges hurt. Unconditional message passing is therefore the wrong default, and reliability-aware fusion the safer one. Every value reported here comes from the accompanying executable code and its saved experiment outputs.

Downloads

Download data is not yet available.

References

[1] Motie, S., & Raahemi, B. (2024). Financial fraud detection using graph neural networks: A systematic review. Expert Systems with Applications, 240, 122156. https://doi.org/10.1016/j.eswa.2023.122156.

[2] Ngai, E. W. T., Hu, Y., Wong, Y. H., Chen, Y., & Sun, X. (2011). The application of data mining techniques in financial fraud detection: A classification framework and an academic review of literature. Decision Support Systems, 50(3), 559–569. https://doi.org/10.1016/j.dss.2010.08.006

[3] Joudaki, H., Rashidian, A., Minaei-Bidgoli, B., Mahmoodi, M., Geraili, B., Nasiri, M., & Arab, M. (2015). Using data mining to detect health care fraud and abuse: A review of literature. Global Journal of Health Science, 7(1), 194–202. https://doi.org/10.5539/gjhs.v7n1p194

[4] Bauder, R. A., & Khoshgoftaar, T. M. (2018). The detection of Medicare fraud using machine learning methods with excluded provider labels. In Proceedings of the 31st International Florida Artificial Intelligence Research Society Conference (FLAIRS-31) (pp. 404–409).

[5] Johnson, J. M., & Khoshgoftaar, T. M. (2019). Medicare fraud detection using neural networks. Journal of Big Data, 6, 63. https://doi.org/10.1186/s40537-019-0225-0

[6] Yoo, Y., Shin, D., Han, D., Kyeong, S., & Shin, J. (2022). Medicare fraud detection using graph neural networks. In 2022 International Conference on Electrical, Computer and Energy Technologies (ICECET) (pp. 1–5). https://doi.org/10.1109/ICECET55527.2022.9872963

[7] Muhammad, R., Tbaishat, D., Nazir, A., Yacoub, S., AbdulRazek, M., Abo El-Enen, M. A., & Sahlol, A. T. (2025). Fraud detection and explanation in medical claims using GNN architectures. Scientific Reports, 15, 41734. https://doi.org/10.1038/s41598-025-22910-6

[8] Liang, C., Liu, Z., Liu, B., Zhou, J., Li, X., Yang, S., & Qi, Y. (2019). Uncovering insurance fraud conspiracy with network learning. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval (pp. 1181–1184). https://doi.org/10.1145/3331184.3331372

[9] Chen, C., Li, Z., Sun, X., Tian, G., Jiang, J., Zhang, K., & Wang, H. (2019). InfDetect: A large scale graph-based fraud detection system for e-commerce insurance. In 2019 IEEE International Conference on Big Data (Big Data) (pp. 1765–1773). https://doi.org/10.1109/BigData47090.2019.9006115

[10] Kipf, T. N., & Welling, M. (2017). Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations.

[11] Hamilton, W. L., Ying, R., & Leskovec, J. (2017). Inductive representation learning on large graphs. In Advances in Neural Information Processing Systems 30 (pp. 1024–1034).

[12] Veličković, P., Cucurull, G., Casanova, A., Romero, A., Liò, P., & Bengio, Y. (2018). Graph attention networks. In International Conference on Learning Representations.

[13] Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., & Yu, P. S. (2021). A comprehensive survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, 32(1), 4–24. https://doi.org/10.1109/TNNLS.2020.2978386

[14] Dou, Y., Liu, Z., Sun, L., Deng, Y., Peng, H., & Yu, P. S. (2020). Enhancing graph neural network-based fraud detectors against camouflaged fraudsters. In Proceedings of the 29th ACM International Conference on Information and Knowledge Management (pp. 315–324). https://doi.org/10.1145/3340531.3411903

[15] Liu, Y., Ao, Z., Qin, Z., Chi, J., Feng, J., Yang, H., & He, Q. (2021). Pick and choose: A GNN-based imbalanced learning approach for fraud detection. In Proceedings of the Web Conference 2021 (pp. 3168–3177). https://doi.org/10.1145/3442381.3449989

[16] Huang, M., Liu, Z., Sun, Y., Zhu, L., Yu, P. S., & He, Q. (2022). AUC-oriented graph neural network for fraud detection. In Proceedings of the ACM Web Conference 2022 (pp. 1311–1321). https://doi.org/10.1145/3485447.3512178

[17] Lin, T.-Y., Goyal, P., Girshick, R., He, K., & Dollár, P. (2017). Focal loss for dense object detection. In Proceedings of the IEEE International Conference on Computer Vision (pp. 2980–2988). https://doi.org/10.1109/ICCV.2017.324

[18] He, H., & Garcia, E. A. (2009). Learning from imbalanced data. IEEE Transactions on Knowledge and Data Engineering, 21(9), 1263–1284. https://doi.org/10.1109/TKDE.2008.239

[19] Davis, J., & Goadrich, M. (2006). The relationship between precision-recall and ROC curves. In Proceedings of the 23rd International Conference on Machine Learning (pp. 233–240). https://doi.org/10.1145/1143844.1143874

[20] Lundberg, S. M., & Lee, S.-I. (2017). A unified approach to interpreting model predictions. In Advances in Neural Information Processing Systems 30 (pp. 4765–4774).

[21] Sundararajan, M., Taly, A., & Yan, Q. (2017). Axiomatic attribution for deep networks. In Proceedings of the 34th International Conference on Machine Learning (pp. 3319–3328).

[22] Ying, Z., Bourgeois, D., You, J., Zitnik, M., & Leskovec, J. (2019). GNNExplainer: Generating explanations for graph neural networks. In Advances in Neural Information Processing Systems 32.

[23] Luo, D., Cheng, W., Xu, D., Yu, W., Zong, B., Chen, H., & Zhang, X. (2020). Parameterized explainer for graph neural network. In Advances in Neural Information Processing Systems 33 (pp. 19620–19631).

[24] Yuan, H., Yu, H., Gui, S., & Ji, S. (2023). Explainability in graph neural networks: A taxonomic survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(5), 5782–5799. https://doi.org/10.1109/TPAMI.2022.3204236

[25] Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). Why should I trust you?: Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1135–1144). https://doi.org/10.1145/2939672.2939778

[26] Zhou, T. (2024). Healthcare Provider Fraud Detection Analysis. Mendeley Data (Version 1). https://doi.org/10.17632/552r52kmgb.1

[27] Hong, B., Zhang, Y., Li, Y., Wang, Y., & Zhao, L. (2024). Health insurance fraud detection based on multi-channel heterogeneous graph structure learning. Heliyon, 10, e30045. https://doi.org/10.1016/j.heliyon.2024.e30045

[28] Chen, J.-P., Lu, P., Yang, F., Chen, R., & Lin, K. (2022). Medical insurance fraud detection using graph neural networks with spatio-temporal constraints. Journal of Network Intelligence, 7(2), 480–498.

[29] Lu, J., Wang, Y., Zhang, Y., Li, Y., & Zhao, L. (2023). Health insurance fraud detection by using an attributed heterogeneous information network with a hierarchical attention mechanism. BMC Medical Informatics and Decision Making, 23, 62. https://doi.org/10.1186/s12911-023-02152-0

[30] Zhang, J., Yang, F., Lin, K., & Lai, Y. (2022). Hierarchical multi-modal fusion on dynamic heterogeneous graph for health insurance fraud detection. In 2022 IEEE International Conference on Multimedia and Expo (ICME) (pp. 1–6).

[31] Hancock, J. T., Bauder, R. A., Wang, H., & Khoshgoftaar, T. M. (2023). Explainable machine learning models for Medicare fraud detection. Journal of Big Data, 10, 154. https://doi.org/10.1186/s40537-023-00821-5

[32] Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). SMOTE: Synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 16, 321–357. https://doi.org/10.1613/jair.953

[33] Jiao, Y., Fan, H., Yue, X., Ping, W., Sun, T., & Wang, J. (2026). Dynamic heterogeneous graph contrastive learning for uncovering collusive financial fraud. Scientific Reports, 16, 12456. https://doi.org/10.1038/s41598-026-58938-5

[34] Fan, H., Jiao, Y., Wang, M., Chen, J., & Yang, S. (2026). Fraud learns too: Continual graph learning under strategic adversarial drift in dynamic networks. Scientific Reports, 16, 14021. https://doi.org/10.1038/s41598-026-60997-7

[35] Ping, W., Jiao, Y., Fan, H., & Zhang, X. (2026). Multimodal fraud detection in financial statements: A trimodal attention network with contrastive evidence chain construction. IEEE Access, 14, 80456–80468. https://doi.org/10.1109/ACCESS.2026.3695466

[36] Chen, J., Liu, J., Liang, Y., & Zhou, M. (2026). KE-MLLM: A knowledge-enhanced multi-sensor learning framework for explainable fake review detection. Applied Sciences, 16, 2909. https://doi.org/10.3390/app16062909