I try to first use regression to get the value of variable V, then use V to discriminate sample A and B.
The distribution of regression target V shows good separation between sample A and B.
If I use only A to train the regression, the regressed value RegV in A and B are both similar to target V in A.
So in order to keep the information that target V are different in A and B, I merge A and B and use A+B to train the regression, the RegV in A and RegV in B are both now similar to target V(A+B), so that the separation power decrease compared to that of target V shown between sample A and B.
I’m wondering why this happens?
Is there any solution?
Sincerely thanks in advance,