Automated and Self-Adapting Approach to AI-based Anomaly Detection
Sheng Ding, Adrian Wolf, Andrey S. Morozov · 2023
Time series anomaly detection (TSAD) is vital across industries, helping identify abnormal patterns to prevent issues, reduce costs, and improve system performance.Nowadays, AI has emerged as a promising solution to enhance TSAD.However, it lacks the self-adapting ability and knowledge of choosing the best-suited model under different contexts.To overcome these challenges, we have integrated various algorithms using a unified data interface and an automated training-testing process.We have incorporated automated hyperparameter optimization and architecture selection.Additionally, we conducted further experiments that demonstrated the advantages of a smart switch mechanism for selecting the most appropriate TSAD method based on statistical features of the data, resulting in improved detection performance.This dynamic switch mechanism has been integrated into our TSAD platform.