Mamba-Based Temporal Feature Learning for Anomaly Detection in Distributed Microservice Environments

Abstract

To address the challenges of identifying abnormal behavior, strong temporal correlations in runtime states, and complex dynamic changes in multidimensional monitoring data within backend microservice systems, this paper proposes an anomaly detection method based on the Mamba state-space model. This method focuses on the continuously generated monitoring sequences during microservice operation, constructing a unified detection framework around temporal state modeling and anomaly representation learning to enhance the ability to identify complex anomaly patterns. First, the raw monitoring data is serialized and feature-mapped, transforming heterogeneous runtime signals into a compact representation suitable for temporal modeling. Then, a gating modulation mechanism is introduced to adaptively filter key information in the input features, enhancing the expression intensity of anomaly-related patterns in the feature space. Based on this, a selective state transition mechanism is used to model the evolution of hidden states under continuous input, thereby more effectively preserving long-range temporal dependencies and perceiving local anomaly disturbances. Simultaneously, a residual enhancement strategy is combined to stabilize the deep representation propagation process, further improving the model's ability to characterize complex runtime states and discriminate anomalies. This research focuses on anomaly data in open-source microservices and evaluates the proposed method against several related approaches. The results show that the proposed method achieves superior performance in terms of precision, recall, F1 score, and AUC, enabling more accurate identification of potential anomalies in backend microservice scenarios. The study demonstrates that the state-space modeling-based anomaly detection framework provides a more effective technical path for anomaly identification in backend microservice systems and offers methodological support for intelligent monitoring and anomaly analysis in complex distributed software environments.