Supervised learning data prep for sport prediction

Hi All,

I've got my sports data and looking to build a model for predicting the outcome of sport games.

My data has a single row for each team ie.

Team StatA StatB StatC
  A      4        6        8       
  B      3        9        8
  C      4        6        5
  .....

Now in my data team A plays team B, Team C plays Team D etc...

Now there's two ways I can do this, the first is a matched pairs (Can rapidminer do this?) so my data would look like

    ID    Team StatA StatB StatC
  G1      A      4        6        8       
  G1      B      3        9        8
  G2      C      4        6        5
       
Then you tell the program ID is the matched ID field so it knows the first row and second row is a matched pair and builds the model accordingly

Or the other way is to transform the data into one row like this,

AStatA AStatB AStatC HStatA HStatB HStatC
    4          6          8        3            9        8 

So now all my data from both teams in a single match is on a single row, and build the model this way.

Can I get pros and cons for each? will it yield the same result and is the first matched pair even possible? (I know it is in SAS)