From a Q&A with a VISA fraud pre­ven­tion agent on reddit:

Some years ago, some­one wrote a paper claim­ing he could get the age, gen­der and race only from the credit card pur­chase his­tory. It worked very well. Today, with your full pur­chase infor­ma­tion, we can even “guess” your income range, num­ber of depen­dants and even weight. We have a sta­tis­ti­cal pro­file of every cus­tomer. We can even cal­cu­late the odds you eat at McDonald’s today, con­sid­er­ing you ate there once every X days. 98% of the time, this model is very accurate.

One draw­back is that it requires a lot of infor­ma­tion. That is why it takes a few years and then, we are fully able to track you. In many cases, we com­pare the pro­file cal­cu­lated from your pur­chase his­tory to who you really are (and you thought they asked your income for credit val­i­da­tion) to fur­ther improve our mod­els, and track fraud, most of all. It’s so sophis­ti­cated that if you order prod­ucts a per­son in your group never ordered, your card will get auto­mat­i­cally locked.

As I’ve men­tioned pre­vi­ously, “I’ve always loved read­ing and learn­ing about data min­ing and its applications”—this is no exception.

From this Q&A I also dis­cov­ered the Luhn algo­rithm.