The main problem I see is the approach of using big win, little win, draw, little loss big loss based on goals scored misses the key point for me:
A big win against Burnley cannot be as significant as a big win against Chelsea.
Researching the problem I found it had been approached in a strategy called RateForm.
The rateform model given in that article is a generalization of the ELO system used in chess and I was lucky enough to have some old code for calculating chess ratings based on an article by Vincent Bisset of Malahide Chess Club on the Irish Chess Union website.
There are also some recommedations as to how this system could be tweaked to cater for home bias in this article from chessbase.
So after installing Visual Studio 2010 and taking Silverlight for a spin I now have an application that rates teams performance over time based on results against the opposition they play.
The screenshot shows the output for the past year in hard hitting Rugby League Football.
Now I am going to see how this plays into the fuzzy model.
