Hands-On Artificial Intelligence for Cybersecurity

Supervised learning pros and cons

However logical it may seem to follow a supervised learning strategy, it is necessary to take into consideration the methodological risks involved.

One of the main problems is that our algorithms will have difficulty in learning to recognize new cases of suspicious activity as they will have been conditioned by previous classification labels, which may be affected by systematic errors. In order to proceed with the retraining of our models, to detect new forms of suspicious activity, we would be forced to insert different classification rules from the previous ones, which should be able to correctly detect the new labels associated with new samples.

However, this does not prevent the risk of amplifying the ...