model from an OLAP cube, too many cases evaluated...

Hi,

I create a model from a cube and the case attribute contains 2.6millions of
members.
but for the specific cube I'm working on for my model contains data for only
1millions of members.

but the DM model continue to use and analyze 2.6m so most of the time my
prediction is "non existing".

So how I can insure that the model will use only members really used in my
source cube instead of all the members from the dimension?
I try some "not null" options and filters, but nothing change.

Thanks for your help.

Jerome.