Learning or Bayesian Statistics are used by MailWasher to help recognize spam and good email, based on email you train as spam or good. After a short period of training, the learning filter becomes very accurate.
In the Inbox screen, the column Classification shows thumbs up or thumbs down. Clicking the thumbs up icon means you think the email is good, while clicking the thumbs down icon means you think the email is spam. Clicking a colored thumbs up or down icon again will cause it to change to neutral.
These actions of determining if an email is good, spam or neutral help build up a body of good and spammy words inside emails. To begin with after training only a few emails, the accuracy may not be great, but after training around 20 emails of good and spam content, the accuracy should be very high. As you train more emails, the accuracy becomes higher and the training will largely take care of itself.
In Settings>>Spam Tools>>Learning there are various options for changing the settings, but consult the Bayesian Technical Information first.