Fraudulent behavior can have multiple forms and is constantly evolving. Someone with bad intentions might steal a credit card and transfer a large amount of money to another account. This kind of transaction can be identified with traditional statistical methods and/or machine learning. The goal of these algorithms is to spot anomalies, rare events that do not correspond to the normal expected pattern; for instance, if your credit card starts to be used from another country than your usual place of living, it would be highly suspicious and likely to be flagged as fraudulent. This is the reason why some banks will ask you to let them know when you are traveling abroad so that your credit card does not get blocked.
Imagine a criminal that withdraws $1 from one billion credit cards, for a total theft of $1 billion, instead of one single person transferring $1 billion from one card at once. While the latter case can be identified easily with the traditional methods...
