Artificial Intelligence for Cybersecurity

When working with supervised learning problems, we are provided with a dataset already labelled (we have some past data about a phenomenon and its resulting outcome). In this scenario, we can be either interested to learn how to predict to which class a new data point might belong to (classification) or to assign it a numerical value from a continuous spectrum (regression). Regression problems could be used in cybersecurity for example in order to try to predict how many devices could have been corrupted by a cybersecurity attack or quantify the level of damage due to a security attack. An example of a classification problem could instead be trying to predict if a login attempt is genuine or not (based on factors such as location, time, machine, etc.). When working with classification problems it can be extremely important to make sure to have a balanced number of examples for each possible output class.