Graph Neural Networks for Cross- Domain Failure Correlation Across Pipelines, ML Models, and Analytics SLAs in Retail Enterprises

Abstract

Retail enterprises often monitor pipelines, ML models, and analytics SLAs in separate observability layers, which makes it difficult to recognize when failures are part of the same operational incident. Existing anomaly-detection and observability methods improve visibility within individual domains, but crossdomain failure correlation remains limited when alerts are analyzed in isolation. This article presents a graph neural network framework that connects pipelines, ML services, and SLA entities in a shared operational graph to identify correlated failures, propagation paths, and incident clusters across retail observability domains. The results show that the proposed approach improves cross-domain failure correlation accuracy, alert linking precision, incident grouping quality, and failure propagation analysis across retail enterprise scenarios. The study shows that graph-based observability can provide a practical foundation for breaking monitoring silos and improving incident understanding across interconnected retail systems.