Machine learning for fraud detection
stripe.com
stripe.com
Peter Thiel writes about how fatal machine learning for fraud detection in his book, "Zero to One".
At Paypal, Max Levchin assembled an elite team of mathematicians to study the fraudulent transfers in detail. Then we took what we learned and wrote software to automatically identify and cancel bogus transactions in real time. But it quickly became clear that this approach wouldn't would either. After an hour or two, the thieves would catch on and change their tactics. We were dealing with an adaptive enemy and our software couldn't adapt in response.
The fraudsters adaptive evasions fooled our automatic detection algorithms, but we found that they didn't fool our human analysts as easily. So max and his engineers rewrote the software to take a hybrid approach: the computer would flag the most suspicious transactions on a well designed user interface, and human operators would make the final judgment as to their legitimacy.
"You insist that there is something a machine cannot do. If you will tell me precisely what it is that a machine cannot do, then I can always make a machine that will do just that!"
-- J. von Neumann
Computers will continue to get better at human things as we continue to get better at understanding how human things work. Look at the recent advances in deep learning. This is using only the most crude approximation of human neurons we can identify and caption images with astounding results. Google currently claims that anything that can be done in 0.1 of a second by a human, they can do as well.
Fraud detection relies heavily on unsupervised learning, and for all of history up until the last few years state of the art unsupervised learning was usually SVD + clustering or some variation on that. The current state of the art, things like deep belief networks, are able to achieve markedly superior results.
Additionally this article seems to imply that they are collected labeled data from customers which should help tremendously in modeling fraud. If even if the labels are a small sample recent advances in semi-supervised learning using deep neural nets is even greater than the advances in unsupervised learning.
While I don't disagree that historically it has been wise to include a human element in fraud detection, I don't believe there is any reason to assume that trend will continue indefinitely into the future.
Okay, on a moonless night, overcast, with no lights, a lot of fog, 200 yards away is .... Bingo, a pretty girl, 5' 4", 34, 19, 34, blond, really sweet, wants to be great as a wife and mommy, about 18! Yup, been doing that for years! Try that Google! My advantage: I have a dedicated, autonomous, peripheral processor just for that task!
> You insist that there is something a machine cannot do. If you will tell me precisely what it is that a machine cannot do, then I can always make a machine that will do just that! - Know anything about the Halting problem & NP-Hard? There are things that computers will never-ever be able to perform, even if we go quantum.
> Computers will continue to get better at human things as we continue to get better at understanding how human things work
Your argument is the basic concept of symbolism, and deep learning is one of the multiple connectionism types of learning, a whole different world in ML.
> Fraud detection relies heavily on unsupervised learning
No, i have been working in Fraud and no, it is supervised with lot of manual feedback.
> unsupervised learning was usually SVD + clustering or some variation on that. The current state of the art, things like deep belief networks, are able to achieve markedly superior results.
Sorry, No Free Lunch for learning algorithms...
> historically it has been wise to include a human element in fraud detection, I don't believe there is any reason to assume that trend will continue indefinitely into the future.
Yes, and it will.
Stripe has no relationship with the end user, and should not. A legitimate buyer can't possibly be expected to call into Stripe to verify a transaction before or after a purchase attempt like they could to their own credit card or bank.
(You note that Stripe has no business with the consumer. Well, then, this doesn't affect your relationship with Stripe, because there is none. It affects your relationship with your card's fraud prevention, which has and will always be there...)
And, to be clear, the fact that Stripe is doing fraud protection isn't new—we've always blocked some fraudulent payments, as does every other major payment company. What we're launching is a much better system and, especially, one that businesses using Stripe can train so that there are fewer false positives over time.
A really interesting book that detailed how the development of PayPals anti-fraud system came about (among other things) is Founders at Work[3].
[1] http://www.businessweek.com/stories/2002-09-30/max-levchin-o...
[2] http://www.quora.com/What-were-the-early-achievements-that-d...
if (rand() >= 0.5) {
lock_account()
freeze_balances()
require_many_long_phone_calls()
require_intrusive_amounts_of_scanned_documents()
refuse_to_unlock_account_and_steal_account_balance()
}For example, it's still not possible to report clearly compromised cards to the issuing bank, Visa has no provisions for that. Pretty much tells you all you want to know about their stance on the card fraud.