Best possible performance for your model

Condwras

Hello guys!My name is Jim,this is my second topic here on rapidminer forum!

Great job,i hope we can together find a solution to my problem...!

I am trying to find the best attributes for my model and to take the best possible prediction performance!Wow,i think we all trying to do that,so i will go straight to the point.

I have 35 attributes and i have accuracy(with all 35 atts),32%.

I use the "optimize selection" operator,together with "neural net","apply model" and "performance" operators...All seems to be good and after the process finishes i get my results.On this point i will give you a true example i already working on.When i use all these operators the accuracy i have is about 40%.

Now when i take the attributes that "optimize selection" gives me(this is 13 atts) and try again to measure my performance i dont get 40%.I am going to 35%.

Can anybody explain to me why is that happening?

To be more clear i write after this,the 3 processes i use.....

1)retrieve----x-validation----neural net----apply model----performance
With 35 attributes i get 32%

2)retrieve----optimize selection-x-validation----neural net----apply model----performance
Now the process says to me that i get the best performance is when i use only the 13 attributes.I get a 40% result.

3)retrieve----x-validation----neural net----apply model----performance
Now i use only the 13 attributes as my data...But i get a 35% accuracy....Why is not 40% as i see at step 2?

Thanks a lot and sorry for my bad English.If something is not clear enough,pls ask me for more information

Jim