"One sided performance operator required"
I am trying to predict real number values and I am looking for a performance operator, that is able to distinguish between errors that are higher or lower than the label value.
Imagine you would want to predict the minimum thickness of your ceiling. Everything below the label value would be worthless, the target would be to come as close to the label value as possible but only from above.
is there any way to realize this in Rapid Miner? Ideally I am looking for a rules based performance operator, that I can feed with own rules (somehow similar to the cost performance operator).
Thanks in advance