Tuning the classifiers

In some cases, administrators may want to tune the classifiers. For example, if too many false positives occur, start by setting the sensitivity level to “Narrow.”

It is also possible to combine the classifier with other classifiers, such as looking at certain file types, like both Microsoft Office files and PDF files.

If the overall accuracy level is too low, check to see if all of the positive examples are related to the same subject. If there is a small number of subjects and enough samples for each of them, optionally create a different classifier for each subject:

Steps

  1. Assign a folder to each subject.
  2. Place documents related to the subject in the corresponding folder.
  3. Train the system separately on each folder.

Next steps

In many cases, several small specific classifiers can provide better accuracy than one general classifier.