Definition of Lift

B_Miner

Hi All,

Can anyone explain how RM is defining "lift" as it pertains to the binomial classifer performance operator?

I am used to lift, in this sense, as being the proportion of the top X% of cases sorted by confidence/probability descending, divided by the overall proportion of positive cases in the data. I cant find how you would set the X% in RM. Is my definition correct and is a X% set (non changeable)?

Thanks!

land

Hi.
It is defined as:

lift = x/y with
x = counter / (counter + counter);
y = (counter + counter) / (counter + counter + counter + counter);
and first index being the true label, and the second the predicted.

Greetings,
  Sebastian