We’ve got a big week coming up, with a full schedule of games on Saturday and Sunday and then again in mid-week. Just this weekend’s predictions for the moment and then I’ll try to get a post up on Sunday or Monday for the next set. Partly, we’ll need to adjust for any injuries that come up on Saturday and Sunday so that the starting line-ups in the model for the mid-week round will be as accurate as possible.
I’m continuing to tune the model with a bit of cleaning up around the edges, some work to make sure that ‘big chances’ (as defined by Opta) get converted at the correct rate and also adjusting the length of matches to cope with Arsenal’s and Man City’s (in particular) very high passing frequencies. All together it should be worth an extra 2%, or 5-7 correct predictions across a season. Once you’ve got a reasonable model, you have to fight hard for every tiny improvement!
I hope there’ll be a larger update at some point this season and have quite a few upgrades on the list left to try, but every time I make a change, I simulate a full couple of seasons to see if it’s made things better or worse, so it can be a bit of slow process making improvements.
Before we dive into this week’s predictions, last week’s games were interesting. We had a modest win at the bookies, but overall I was pleased with the model’s forecasts. Here they are as a reminder…
Running through in order, Everton v Liverpool was a brilliant but crazy game that either team could have won. Maybe the model leaned too far towards Liverpool but if Allen had shot with his eyes open, we’d probably have got that one.
Arsenal beat Southampton, Newcastle beat Norwich, Stoke beat Sunderland and West Brom drew with Villa. Lovely.
Of the losing bets, Fulham to draw, Hull to draw and Man U to win were all close. With the level of randomness in football, you’re going to be wrong a lot, but it’s worrying when your predictions are nowhere near. That did happen, in West Ham vs. Chelsea where although the model picked the right winner its draw percentage was high, leading to a losing bet. That’s the single call from last week I’m really not happy about and will be investigating a bit further.
The call I’m delighted with was Man City’s big win over Tottenham. A few eyebrows were raised last Friday over the model’s 5-1 “most likely score” prediction, but it ended up being pretty accurate! Man City’s shooting accuracy at home is spectacular so far this season, but the model also adjusts opponents based on who they’re playing, which led to some poor expected individual performances for Spurs. It certainly won’t predict that everybody gets battered by Man City at home and didn’t for Swansea this week. Well, that was up until those tweaks I mentioned earlier. Now on the other hand…
Here are this week’s percentages:
And like last week, predicted possession and shooting stats for each game:
If you’re betting, I’ll be going for:
Aston Villa vs Sunderland – Home win
Cardiff City vs Arsenal – Away win
Everton vs Stoke City – Home win
Norwich City vs Crystal Palace – Home win
West Ham United vs Fulham – Draw
Newcastle United vs West Bromwich Albion – Home win
Tottenham Hotspur vs Manchester United – Away win
Hull City vs Liverpool – Away win
Chelsea vs Southampton – Home win
Manchester City vs Swansea City – Home win
Only one draw on the list this week and I think that’s unlikely – there will probably be a few – but am following a strategy which has worked overall in the past. If you’re using the model, I’d love to hear about it, either in the comments (the EPL Index ones) or on Twitter using #EPLModel. See you in a few days!