Detecting Microservice's Architectural Anti‐Pattern Indicators Using Graph Neural Networks

ABSTRACT Introduction Organizations moving from monolithic to microservice architectures face new challenges due to distributed complexity. Architectural Anti‐Patterns (Smells) can arise and contribute to Technical Debt, while current detection approaches remain manual or semi‐automated and prone to error. Methods This study aims to address this gap by proposing an automated detection tool leveraging graph neural networks (GNNs). Microservice systems are modeled as graphs, with services as nodes and their relationships as edges. Graph neural networks (GNNs) are applied to detect four anti‐patterns: (1) cyclic dependencies, (2) enterprise service bus (ESB) usage, (3) microservice greediness, and (4) inappropriate service intimacy. Large language models (LLMs) are used to generate and expand architectural graph datasets to address data scarcity. Results The GNN approach achieves improved detection performance, with up to a 1.2% F1‐score increase over existing tools such as Msanose and Arcan. Conclusion Combining GNNs with LLM‐augmented data enhances automated detection of microservice anti‐patterns and supports more effective architectural assessment.