Date: Sun, 3 Dec 2006 22:21:47 -0800
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: Credit Card Fraud Detection
In-Reply-To: <1164464015.434103.178470@14g2000cws.googlegroups.com>
Content-Type: text/plain; format=flowed
piyushk.gupta@GMAIL.COM wrote:
>Hi all,
>
>I am looking to work on a credit card fraud detection model (using
>SAS). The approach will be to develop business rules which when applied
>to every transaction, will give some sort of a risk score, which can be
>used to identify the likelihood of the transaction being good or bad.
>If anybody has any study material/pointers/knowledge on credit card
>fraud detection (or fraud detection in general), your help will be
>highly appreciated.
There is a huge amount of research in this area. Some of it is proprietary,
which means you have absolutely no shot at ever seeing what the
developers did, or what models they use. But I would not recommend
that you try building one yourself from scratch, unless your company can
afford a few years of losing millions of dollars due to sub-standard fraud
detection while you fiddle with your model and keep improving it and
keep learning about this field.
If it is worth the money to invest in fraud detection, it is worth the
money to invest in already-built fraud detection systems. Contact
a SAS rep and ask him/her about SAS solutions in this area.
>Also, I already tried to develop one such model with some credit card
>transaction datasets that I had access to. However, the problem was
>that because they werent meant to be used for fraud detection, the data
>sets did not have a labelled fraud field, ie, one which says whether
>the transaction was good or bad. For me, without that field it was
>impossible to build a meaningful model. Can anybody think of a
>workaround - like generating a dummy fraud field for example? Maybe I
>didnt have enough knowledge about building such systems. Your inputs
>will be really helpful.
There is no way to do this. Period.
Any fake fraud variable will only lead to a model which looks at that
variation, and which has nothign to do with real fraud detection. You
need real data to build such a model.
>Thanks,
>Piyush.
>
>PS: Apologies for cross posting.
Then why did you do it?
HTH,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330
_________________________________________________________________
Get free, personalized commercial-free online radio with MSN Radio powered
by Pandora http://radio.msn.com/?icid=T002MSN03A07001
