I'm studying this article. The authors used a two-step approach to detect anomalies. First, they used an LSTM to learn the normal behavior of the data. Then, they used the dynamic error thresholds to detect deviations from the normal behavior. I have some troubles with understanding the dynamic thresholding section. For instance I don't understand what are the following items:

  1. epsilon and epsilon
  2. E_seq and |E_seq^2|
  3. What is p which is explained in Figure 2.

Can anyone help me to understand this algorithms better?

  • $\begingroup$ p is the minimum percentage decrease which is used to determine wheter or not a detected anomaly is actually kept as an anomaly or reclassified as a normal observation. This is compared to the percentage difference between the detected anomalies (or to the previous highest error in case of the first detected anomaly). $\endgroup$
    – Oxbowerce
    Aug 4, 2023 at 17:00


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.