Hi! i am a junior dev need advice regarding fraud\/risk scoring (not credit) on my rules based fraud detection system.
\n \t
\n
<\/BR>
\n
\n \t
\n
<\/BR><\/p>\n
so i our team has developed a rules based fraud detecton system….now we have received a new requirement that we have to score every transaction as how much risky or if flagged as fraud how much fraud it is.<\/p>\n
i did some research and i found out its easier if it is a supervisied operation but in my case i wont be able to access prod transaction data due to policy.<\/p>\n
now i have 2 problems data which i guess i have to make a fake one.<\/p>\n
2nd how to score i was thinking of going witb regression if i keep my target value bete 0 and 1 but realised that the model can predict above that then thought of classification and use predict_proba() to get prediction probability.<\/p>\n
or isolation forest<\/p>\n
till now thats what i bave you thought what else shoudl i consider any advices or guidance to set me in the right path so i dont get any rework<\/p>\n<\/p><\/div>\n
submitted by \/u\/1_plate_parcel <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span>\n<\/div>\n