Authors:
Robson S. Santos
1
;
Robesvânia Araújo
1
;
Paulo Rego
1
;
José M. da S. M. Filho
1
;
Jarélio G. da S. Filho
2
;
José D. C. Neto
2
;
Nicksson C. A. de Freitas
2
;
Emanuel Rodrigues
1
;
Francisco Gomes
1
and
Fernando Trinta
1
Affiliations:
1
Federal University of Ceará (UFC), Av. Humberto Monte, s/n, Pici – 60440-593 – Fortaleza – CE, Brazil
;
2
Sidi – Institute of Innovation for Digital Society, Av. República do Líbano, 251 – 51110-160 – Recife – PE, Brazil
Keyword(s):
Fraud Detection, Financial Systems, Microservice Architecture, Machine Learning, Observability, Scalability.
Abstract:
The growth of real-time financial transactions has increased the demand for scalable and transparent fraud detection systems. This paper presents a microservice-based architecture designed to detect credit card fraud in real time, integrating machine learning models with observability tools to monitor operational behavior. Built on OpenTelemetry (OTel), the architecture enables detailed tracking of performance metrics, resource usage, and system bottlenecks. Experiments conducted in a cloud-based environment demonstrate the scalability and efficiency of the solution under different workloads. Among the tested models, XGBoost outperformed Random Forest in throughput and latency, handling over 25,000 concurrent requests with response times under 50 ms. Compared to previous work focused solely on model accuracy, this study advances toward real-world applicability by combining fraud detection with runtime observability and elastic deployment. The solution is open-source and reproducible,
and it contributes to the development of robust data-driven systems in the financial domain.
(More)