Hands-On Artificial Intelligence for Cybersecurity

Dealing with non-stationary datasets

In order to manage the non-stationary characteristic of the distribution, it may be useful to overweigh the feedback that was obtained by human operators, which contributes to improving the classification of supervised samples.

Therefore, in the presence of non-stationary data, it may be useful to use an ensemble of classifiers (ensemble learning), whose training is carried out on different samples, to improve the overall prediction accuracy.

By integrating different classifiers, it is possible to combine the knowledge that was obtained on the basis of the new observations with the knowledge that was previously acquired, weighing each classifier on the basis of its classification capability, and excluding ...