Hands-On Artificial Intelligence for Cybersecurity

Sampling methods for unbalanced datasets

A final aspect to consider before moving on to the operational phase of fraud detection relates to the management of unbalanced data.

We have already said that one of the characteristics of credit card transactions is to show unbalanced distributions toward genuine transactions.

To manage this asymmetry in the data, we can use different sampling methods that intend to rebalance the transaction dataset, thereby allowing the classifier to perform better.

The two most adopted sampling modes are undersampling and oversampling. Through undersampling, some random samples are removed from the most numerous class (in our case, the class of legitimate transactions); with oversampling, synthetic samples are ...