Round one of the results prediction model went ok, so here we go with my numbers for Saturday 9th and Sunday 10th November.
I’ve been making a few small changes to the model based on last week – nothing too drastic, but it’s encountering some new problems with shortages of data that I didn’t have last season. I first started building these models in the second half of last season so there were always at least 20 games worth of data to work with. Only ten games in, some of the more volatile stats (see last post) needed to be moderated and I’ve also added a slightly better routine for (gu)estimating the shooting accuracy of strikers who haven’t taken many shots yet.
Some early season problems aren’t completely solveable. Fantasy Football Scout is predicting starts for a few players who we haven’t seen play yet, so I’ve had to take a couple of liberties with the team selections. Dejagah for Fulham is an example – he hasn’t played an away game yet, so we don’t have any performance numbers for him.
Enough of my data shortage issues. Here are this week’s percentages.
The percentages are much closer to the bookies odds this week, which is a result of moderating those volatile early season stats. Crucially, I’ve capped the highest quality a defence can have at a sensible limit, based on the last five seasons. A few teams – Southampton especially – haven’t conceded many goals yet and based on that the model was trying to say that they almost never would. They’ve still got a strong defence in the model, just not an invincible one.
If you’re betting, please do it with caution for all the reasons I’ve just covered! Here are my calls:
Villa v Cardiff – Draw
Chelsea v West Brom – Home win
Palace v Everton – Away win
Liverpool v Fulham – Home win
Southampton v Hull – Draw
Norwich v West Ham – Away win
Spurs v Newcastle – Home Win
Sunderland v Man City – Away win
Man U v Arsenal – Away win
Swansea v Stoke – Draw
A couple of those are worth some quick discussion. Firstly, the model’s obviously got over its Fulham favouritism from last week and is predicting an easy win for Liverpool.
Arsenal’s win percentage is very high, because the model is still being strongly affected by Man U’s first few games and the fact that Man U have conceded a lot more goals than they normally would, given the number of shots they’ve faced. Plus the structure of the model will tend to favour teams that pass well and dominate possession. Again, a bit more data will settle this type of call down.
Fingers crossed and see you next week for a results round-up.