Anomaly Detection in Microservices Systems via Call Chain Analysis and Machine Learning
Jing Du, Meng Yang, Yue Shao, Feng Zhang, Sihan Tian · Advances in transdisciplinary engineering · 2025
Microservices architecture, with its scalability and modularity, has become the foundation for modern distributed systems. However, the complexity of service interactions, often represented as intricate service call chains, creates challenges in identifying anomalies accurately and efficiently. This paper introduces an innovative framework for anomaly detection in microservices-based systems, integrating call chain analysis with machine learning techniques. The framework leverages distributed tracing data to construct Service Dependency Graphs (SDGs) and extract critical features, including temporal metrics such as latency and topological metrics like dependency weights. These features are processed by advanced machine learning models, including Isolation Forests and Long Short-Term Memory (LSTM) networks, to detect anomalous patterns in real time. Experimental results on a cloud-native microservices platform demonstrate the framework’s high performance, achieving an F1-score of 94.3% while reducing detection latency by over 70% compared to traditional rule-based methods. Additionally, the framework exhibits robust scalability, maintaining low detection delays even under increasing service complexity. This work provides a practical and scalable solution for enhancing anomaly detection in dynamic and large-scale microservices environments.