Next-Gen Cloud Healthcare ERP Security Framework Leveraging Multivariate Classification, BERT, and Databricks for Real-Time Staffing and Risk Analytics

Abstract

In modern financial systems, fraud detection demands high scalability, near real‑time processing, and agile evolution of data storage structures. Legacy relational databases — often rigid and monolithic — impair the capability to rapidly evolve schema, integrate machine‑learning (ML) pipelines, and support continuous delivery. This paper presents a cloud‑native database upgrade architecture that embeds a GitHub‑automated CI/CD pipeline, supports generic resource abstraction (GRA) for database schema evolution, and integrates machine‑learning driven fraud detection workflows. The architecture enables safe, version-controlled schema migrations, supports ML model deployment and retraining, and ensures data integrity and minimal downtime. We describe the design, implementation, and evaluation of this architecture in the context of large-scale fraud detection pipelines. Empirical evaluation shows that schema changes, ML model updates, and data migrations can be automated through CI/CD with rollback safety, reducing deployment time by over 60%, minimizing schema‑drift incidents, and allowing near real-time fraud detection on streaming transaction data. Further, ML models trained within the pipeline achieved high detection accuracy with low false‑positive rates, and retraining cycles could be deployed with minimal operational overhead. The proposed architecture thus bridges database DevOps practices with ML operations, enabling financial institutions to respond rapidly to evolving fraud patterns while maintaining rigorous control over schema evolution and data consistency.