I am looking for a technical understanding for detecting a "Univariate Categorical Outlier".
I have used the ML Toolkit on Splunk and basically, I am trying to detect the "rare" categories which are really having low frequencies for the given variable of the dataset.
I have also followed the thread here but I couldn't find the information I am looking for. Tough I could see the links like this which discuss different methods like histogram, IQR, and ZScore for anomaly detection but couldn't find any technical overview.
If anyone could help me with finding the "rare" category automatically, it will be a huge help. Because setting a static threshold like 0.05 doesn't work for all datasets. There has to be some way around like the histogram method.
Please give me the sources on how splunk finds the rare categories. It is fine if you can provide me with the univariate variable only instead of the multivariate.
Thanks
