In Belgium 25% of the games do end up in a draw. Almost the same figure for the games in the Premier League.
The weakness of the Poisson models is that on one side it most of the time does predict low scoring games, hence to many 0-0, 1-0 or 0-1 s are predicted.
Also it rarely does predict a draw! Besides the most likely correct score.
How do you adopt the predictions with the weight. Very curious about it.