Posts

Showing posts from February, 2020

Anomaly Detection and Change Point Detection - Reproduced

Image
Ano malies are patterns in the data that do not conform to a well-defined notion of normal behavior. Techniques used to detection anomalies typically require training before using on new data. Here we will reproduce the results from  Oana Niculaescu 's article in  XRDS ,  Applying Data Science for Anomaly and Change Point Detection . This article was generated using a Jupyter Notebook. The notebook is available  here . Detecting Changes The  CUSUM  algorithm is used to test for anomalies. This requires two parameters:  threshold  and  drift . But, how do you choose values for these parameters? Gustafsson (2000) provides this recipe: Start with a very large  threshold . Choose  drift  to one half of the expected change, or adjust drift such that  g = 0  more than 50% of the time. Then set the  threshold  so the required number of false alarms (this can be done automatically) or delay for detection is obtained...