Hi,

At work i am trying to correlate fraud with a few other variables . Dont really what I should use:

Scenerio:

I have about 100k Users, out of which 30k have been proven fraud.

I was trying to identify the patterns among the identified fraud users and compare them to the non fraud base.

Eg:

1) avg order value : 1000$ - Fraud user / 600$ - NF 2) avg time spent on website : 20 minutes - FU / 4 minutes - NF 3) Avg categories shopped: 1- FU / 3 NF

Now i see a distinct pattern in the 3 aforementioned , how do i correctly correlate them and use them as a measure of identifying a fraud user?

what i was doing:

Assigning random weightages to the above 3 kpis

eg: if avg order value <= 600 then 0 weightage , if its above then higher weightage and so on.

Question:

1) what im doing, is this the right approach? if so, by what logic do i assign weightages instead of randomly? 2) If its not the correct logic, how do i go about using these 3 or any given kpis/ variables and using them to identify a fraud user?

p.s: Example above is hypothetical