Hands-On Artificial Intelligence for Cybersecurity

Learning from unbalanced and non-stationary data

In Chapter 1, Introduction to AI for Cybersecurity Professionals, we saw how machine learning algorithms are divided into supervised and unsupervised learning; this subdivision is also valid in regards to credit card fraud detection, although attention must be paid to the different assumptions that inspire the two categories of algorithms. This is because they have important consequences on the reliability and accuracy of the predictions.

In the case of supervised learning algorithms, it is assumed that a dataset of already categorized samples (labeled samples) is available; that is, each sample was previously associated with one of the two possible categories (legitimate or fraud).