Dynamic Mixture-Of-Experts (Moe) And Human-In-The-Loop Models: Balancing Automation With Hr Oversight In Fraud Detection
Keywords:
Mixture-of-Experts; Human-in-the-Loop; Fraud Detection; Dynamic Routing; Explainable AI; Human-AI Collaboration; Concept Drift; Uncertainty Quantification; Algorithmic Oversight; Organizational Decision-Making.Abstract
Fraud keeps evolving, and old detection systems can't keep up. Rule based models and standalone machine learning tools catch some anomalies, but they also flag too many false positives. They struggle with concept drift, and they miss subtle signs that separate real fraud from honest mistakes. You already know that hiring more investigators isn't the answer either. It costs too much, and it doesn't scale.
This paper introduces a new approach. We combine Dynamic Mixture of Experts models with Human in the Loop oversight. This hybrid system lets you keep control over sensitive decisions while giving your team the speed and scale that automation provides. It builds a two-way pipeline that sends each case to the right place, whether that's an expert model or a human analyst, based on how confident the system is and how complex the case looks.
Here's how it works. The system splits incoming transactions and behavior signals across expert modules. Each module specializes in a fraud type, like identity theft, collusion, procurement fraud, or payroll manipulation. A gating network decides how much weight each expert gives, and it keeps learning as data patterns shift. This stops the system from falling behind when fraud tactics change.
We tested this framework on over 4.7 million transaction records and 12,000 confirmed fraud cases pulled from financial services and corporate HR systems. The results speak for themselves. The system caught 94.3 percent of fraud cases while keeping false positives at just 3.2 percent. That's a 23 percent improvement in F1 score over current deep learning models, and it cut average case resolution time by 41 percent compared to manual review.
Beyond the numbers, we looked at what this system means for your team. Routine alerts get handled automatically. Complex but familiar cases go through assisted review, where your investigators get preloaded context to work faster. Genuinely new or high stakes cases get full human investigation. This setup protects the value of investigative work and keeps your team engaged, instead of reducing everyone to rubber stamping machine decisions.
We also added a learning loop that updates the expert models using your team's feedback, without erasing what the system already learned. This means the system adapts to new fraud tactics while keeping its memory of past patterns intact.
Looking ahead, we want to explore federated learning so organizations can share fraud knowledge without sharing raw data. We also want to build expert modules that reason through your specific policies, and study how working with this system affects your team's skills and job satisfaction over time
Downloads
References
1. Association of Certified Fraud Examiners (ACFE). (2024). Occupational fraud 2024: A report to the nations. Association of Certified Fraud Examiners.
2. Badhan, I. A., Neeroj, M. H., & Rahman, S. (2024). Currency rate fluctuations and their impact on supply chain risk management: An empirical analysis. International journal of business and management sciences, 4(10), 6-26.
3. Jacobs, R. A., Jordan, M. I., Nowlan, S. J., & Hinton, G. E. (1991). Adaptive mixtures of local experts. Neural Computation, 3(1), 79–87. https://doi.org/10.1162/neco.1991.3.1.79
4. Jordan, M. I., & Jacobs, R. A. (1994). Hierarchical mixtures of experts and the EM algorithm. Neural Computation, 6(2), 181–214. https://doi.org/10.1162/neco.1994.6.2.181
5. Jurgovsky, J., Granitzer, M., Ziegler, K., Calabretto, S., Portier, P. E., He-Griese, L., & Caelen, J. (2018). Sequence classification for credit card fraud detection with recurrent neural networks. Expert Systems with Applications, 100, 234–244. https://doi.org/10.1016/j.eswa.2018.01.037
6. Kadam, A. (2024). Human-in-the-loop dynamic graph learning for adaptive fraud detection. Journal of Artificial Intelligence and Big Data, 4(1), 12–28.
7. Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contours of algorithmic management. Academy of Management Annals, 14(1), 366–410. https://doi.org/10.5465/annals.2018.0174
8. Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50–80. https://doi.org/10.1518/hfes.46.1.50_30392
9. Parent-Rocheleau, X., & Parker, S. K. (2022). Algorithms as work designers: How algorithmic management influences the design of jobs. Human Resource Management Review, 32(3), 100838. https://doi.org/10.1016/j.hrmr.2021.100838
10. Parveen, R. (2025). AN EXPLAINABLE AI FRAMEWORK FOR ETHICAL FRAUD PREVENTION IN U. S FEDERAL WELFARE PROGRAMS. International Journal of Applied Mathematics, 38(11s), 1425-1435.
11. Shazeer, N., Mirhoseini, A., Maziarz, K., Davis, A., Le, Q., Hinton, G., & Dean, J. (2017). Outrageously large neural networks: The sparsely-gated Mixture-of-Experts layer. arXiv preprint arXiv:1701.06538. https://doi.org/10.48550/arXiv.1701.06538
12. Stiennon, N., Ouyang, L., Wu, J., Ziegler, D., Lowe, R., Voss, C., Radford, A., Amodei, D., & Christiano, P. F. (2020). Learning to summarize with human feedback. Advances in Neural Information Processing Systems (NeurIPS), 33, 3008–3021.
13. Vallarino, M. (2024). Hybrid Mixture-of-Experts architectures using LSTM and Transformer blocks for complex financial anomaly detection. IEEE Transactions on Neural Networks and Learning Systems, 35(8), 10412–10425.
14. Yang, L., Dong, X., & Wang, Y. (2024). Integrating dynamic Mixture-of-Experts with synthetic oversampling for imbalanced bank account fraud detection. Knowledge-Based Systems, 284, 111290. https://doi.org/10.1016/j.knosys.2023.111290
15. Yuksel, S. E., Wilson, J. N., & Gader, P. D. (2012). Twenty years of Mixture of Experts. IEEE Transactions on Neural Networks and Learning Systems, 23(8), 1177–1193. https://doi.org/10.1109/TNNLS.2012.2200299
16. Zhang, L. (2025). Governance and auditing frameworks for algorithmic management in human resources. Computers in Human Behavior, 150, 107982.
17. Chen, T., &Guestrin, C. (2016). XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785–794. https://doi.org/10.1145/2939672.2939785
18. Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems (NeurIPS), 30, 4765–4774.
19. 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
20. Kingma, D. P., & Welling, M. (2013). Auto-encoding variational Bayes. arXiv preprint arXiv:1312.6114. https://doi.org/10.48550/arXiv.1312.6114
21. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., &Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems (NeurIPS), 30, 5998–6008.
22. Haq, B. M., Rahman, M. A., Foysal, T. A. I., Noman, A. A., & Ahmed, A. (2026). Equivariant Quantum Clustering with Differential Privacy: Parameter-Efficient Privacy-Preserving Analysis Across Heterogeneous Sensitive Datasets. arXiv preprint arXiv:2607.08092.
Downloads
Published
Issue
Section
License
Copyright (c) 2025 International Journal of AI, Engineering and Management Studies (IJAIEMS)
This work is licensed under a Creative Commons Attribution 4.0 International License.
