Unsupervised Anomaly Detection Based on CNN-VAE with Spectral Residual for KPIs
Gongliang Li, Zepeng Wen, Xin Yu Xie · 2022
Current large-scale applications, such as trading systems, blockchain, social software, etc, are increasingly adopting microservice architecture, which bring challenges to manual operation and maintenance, intrusion detection. In both operations and intrusion detection, there are a common characteristic that service metrics and network traffic are normal for most of the time, but anomaly data is more important. In this paper, we propose an unsupervised anomaly detection algorithm based on convolutional neural network with Spectral Residual, which is verified experimentally and has potential application capability with 19.2% f1-score improvement compared to the Variational AutoEncoder.