I need a suggestion about the attribute selection technique in my project

Hi guys,

I just start to play with Weka on my current project.

My data set is a little special. I have some attributes(for example, 20 attributes) and one predicted attribute. However, the 20 attributes are in different files with the predicted attributes. Just give a simple example:

File 1:
A1 A2 A3.... B
1 2 3 .... 5
2 3 4 .... 7
....................

File 2:
A11 A12 A13.... B
11 22 33 .... 50
22 33 44 .... 70
....................

A1-A20 are attributes and B is the predicted one. It's impossible to merge the two files since each instance happens under different situation even when B are the same.
So I want to do some attribute selection to help my classifier (probably I will use one among decision tree, neural network and SVM). So does anyone know how should I do that? What evaluator and search method I should use? I need to compare the attributes among different run.... Currently I am thinking probably Rank is a good choice.:>