AI-Driven Anomaly Detection in Cloud-Native Microservices: The Night’s Watch Algorithm

As organizations shift to microservice architectures, the need for effective anomaly detection becomes more critical. Classic approaches rely heavily on predefined thresholds or labeled data, both of which scale poorly in distributed and dynamic environments. This paper introduces the Night’s Watch algorithm, a novel unsupervised method for detecting anomalies in microservices. By integrating multi-source data and temporal features, the algorithm addresses key limitations of existing approaches. Our experiments demonstrate that the Night’s Watch algorithm significantly improves precision (up to 92%) and recall (up to 39%) depending on the training set size. These results indicate that the algorithm can reduce false positives and enhance real-time anomaly detection in microservice environments. These findings contribute to the development of more robust AI-driven monitoring systems, advancing the state of anomaly detection in cloud-native architectures.