You can beat SOTA Time Series Anomaly Detection methods with a 100 year old algorithm [R]

Excerpt

You can beat SOTA Time Series Anomaly Detection methods with a 100 year old algorithm Time Series Anomaly Detection (TSAD) seems to be one of the hottest topics in NeurIPS, SIGKDD, VLDB etc. Many (perhaps most) papers evaluate on Paparrizos’ TSB-AD-M benchmark… However, I tested these benchmark datasets and found that in most cases I could beat the SOTA TSAD methods with a 100-year-old algorithm, simple Statistical Process Control (SPC). In the attached example, SPC gets perfect results. If we c