Graph-Based Deep Learning for Fault Localization in Service Dependency Networks

Abstract

As microservice architectures scale in complexity, identifying the root causes of failures in distributed service ecosystems becomes increasingly challenging. Traditional fault localization approaches often fall short in capturing the intricate dependency relationships and dynamic behaviors of services. This paper presents a graph-based deep learning framework designed to perform fault localization in service dependency networks with high precision and explainability. By modeling service-to-service interactions as directed graphs and employing Graph Neural Networks (GNNs) to capture structural and temporal patterns, the proposed method outperforms conventional statistical and rule-based techniques. Experimental evaluation on real-world microservice datasets shows that our model can detect and localize faults with significant improvements in accuracy, latency, and robustness. This work lays the foundation for autonomous monitoring and recovery in cloud-native environments.