Deep learning models compared for spotting cloud system anomalies early
Benchmarking Deep Learning Models for Anomaly Detection in Real-World Cloud Telemetry
Machine LearningDistributed, Parallel, and Cluster Computing
Summary
Cloud platforms create huge amounts of data that can be messy and hard to interpret. The authors studied how well different deep learning tools can find unusual patterns or problems in this data quickly. They found that some models catch early signs better but may make more mistakes, while others are more steady but less sensitive. Overall, real cloud problems often look like brief changes rather than ongoing patterns, so it’s important to detect issues fast and carefully handle false alarms.
What this means in practice
- •For cloud platform operators: Improve monitoring by using deep learning models tailored to detect short issues early in cloud service data.
- •For it operations teams: Enhance alert systems by integrating timing-aware evaluation and post-processing to reduce false alarms in anomaly detection.
Authors
Sumit Dhyani, Sougata Mukherjea
Abstract
Modern cloud platforms generate high-dimensional, noisy telemetry from thousands of microservices. Unlike classical anomaly detection benchmarks, cloud anomalies are short-lived, irregular, and largely non-periodic, making timely detection challenging. We present an empirical benchmark of deep learning models for anomaly detection using a real-world IBM Cloud Console dataset. We evaluate GRU, Transformer, TimesNet, and Autoformer architectures using the Numenta Anomaly Benchmark (NAB), a timing-aware metric that accounts for early detection and false alarms. TimesNet achieves the highest NAB scores due to strong early-detection sensitivity. Transformer models also detect anomalies early but incur higher false-positive rates. An improved GRU autoencoder, incorporating robust scoring and persistence rules, achieves competitive performance with a more stable false-alarm profile. Overall, the modest performance differences suggest that real cloud anomalies are dominated by short-term deviations rather than persistent periodic structure, emphasizing the importance of timing-aware evaluation and robust post-processing.