EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

An Agentic AI Framework for Financial Risk Assessment: Integrating Fraud Detection, Explainability, and Privacy Preservation

Dr. Chetan Bulla, Prof. (Dr.) Shashi Kant Gupta, Girish Wali, Anita Kori, Dr. Vinayaksing S Rajaput, Nagaraj Gadagin

Computer and Decision Making: An International Journal · 02.11.2026 · 3 (6S) · 2829-2849

DOI: 10.1922/comdem.v3i6s.19035 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 02.11.2026. Verfasst von Dr. Chetan Bulla und 5 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Financial institutions need intelligent and trustworthy systems that can identify fraud and facilitate financial risk-based decision-making. However, usual AI applications aim at individual risk tasks independently, which makes it challenging to coordinate, explain and make privacy-aware decisions. This paper suggests the Agentic AI framework where the specialized agents for fraud detection, credit-risk assessment, operational risk analysis, anti-money laundering (AML), customer risk profiling and responsible lending can be centralized and orchestrated. The framework consists of long short term memory (LSTM) autoencoders for anomaly detection, generative adversarial networks (GANs) for creating synthetic fraud samples, federated learning to train the model without frequently sharing raw data, and SHapley Additive exPlanations (SHAP) for model interpretation and explainable decision support. The main evaluation is based on the European Credit Card Fraud Detection dataset, a public loan-default dataset and a synthetic e-commerce stream are provided for more general exploration of financial risks and agent coordination. The reported results demonstrate that the receiver operating characteristic area under the curve (ROC-AUC) is equal to 0.90 with a recall equal to 0.75, while the LSTM autoencoder baseline has a ROC-AUC of 0.85 and a recall of 0.50. These findings suggest that these classifiers provide good fraud detection performance and have potential to be used in financial risk decisions with the support of AI. Further validation is required to show cross domain generalization, individual agent effectiveness, privacy assurances and governance outcome

Zitieren

APA

Bulla, D. C., Gupta, P. D. S. K., Wali, G., Kori, A., Rajaput, D. V. S., & Gadagin, N. (2026). An Agentic AI Framework for Financial Risk Assessment: Integrating Fraud Detection, Explainability, and Privacy Preservation. Computer and Decision Making: An International Journal, 3(6S), 2829-2849. https://doi.org/10.1922/comdem.v3i6s.19035

Vancouver

Bulla DC, Gupta PDSK, Wali G, Kori A, Rajaput DVS, Gadagin N. An Agentic AI Framework for Financial Risk Assessment: Integrating Fraud Detection, Explainability, and Privacy Preservation. Computer and Decision Making: An International Journal. 2026;3(6S):2829-2849. doi:10.1922/comdem.v3i6s.19035

BibTeX
@article{Bulla2026,
  author = {Bulla, Dr. Chetan and Gupta, Prof. (Dr.) Shashi Kant and Wali, Girish and Kori, Anita and Rajaput, Dr. Vinayaksing S and Gadagin, Nagaraj},
  title = {An Agentic AI Framework for Financial Risk Assessment: Integrating Fraud Detection, Explainability, and Privacy Preservation},
  journal = {Computer and Decision Making: An International Journal},
  year = {2026},
  volume = {3},
  number = {6S},
  pages = {2829--2849},
  doi = {10.1922/comdem.v3i6s.19035}
}
Kostenlos starten →