Automated Anomaly Detection in Cloud Microservices Using LSTM Autoencoders: A Practical AIOps Approach

In modern cloud-native environments, microservices generate vast amounts of telemetry data, making manual monitoring impossible. Traditional anomaly detection methods, such as static thresholding or statistical rules, often fail to adapt to dynamic workloads, leading to high false-positive rates or missed "silent failures" (e.g., deadlocks, memory leaks). This paper proposes a robust, unsupervised Deep Learning framework utilizing Long Short-Term Memory (LSTM) Autoencoders for AIOps. By training on healthy operational data, the model learns the temporal baseline of server metrics. We introduce a Dynamic Thresholding Mechanism based on the statistical distribution of the Reconstruction Error (μ+3σ) to automatically flag anomalies. The framework is validated using the Numenta Anomaly Benchmark (NAB) on real-world AWS EC2 server logs. Results demonstrate the model's ability to detect complex, non-linear anomalies without prior labeling, offering a scalable solution for automated reliability engineering.