GW 12 Score Predictions

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

GW 12 Score Predictions

SuperGrover
Using current model although I may tweak a bit in the next couple days.  As an fyi I adjusted the home field advantage up to 16%.  this is based upon the past decade of premier league performances.

Once I decide on an additional tweaks for this week I will be publishing an article for the front page.  Until then...

ARS 2.20 - TOT 0.94 (2-1)
LIV 2.29 - WIG 0.85 (2-1)
MCI 2.21 - AVL 0.45 (2-0)
NEW 1.87 - SWA 1.08 (2-1)
QPR 1.92 - SOT 0.98 (2-1)
RDG 1.21 - EVE 1.68 (1-2)
WBA 1.51 - CHE 1.30 (1-1)
NOR 1.43 - MUN 1.93 (1-2)
FUL 2.26 - SUN 0.49 (2-0)
WHM 0.94 - STO 0.96 (1-1)

Reply | Threaded
Open this post in threaded view
|

Re: GW 12 Score Predictions

SuperGrover
This post was updated on .
A couple data updates produced a bit different numbers.  Nothing major:

ARS 2.23 - TOT 0.93 (2-1)
LIV 2.21 - WIG 0.84 (2-1)
MCI 2.28 - AVL 0.47 (2-0)
NEW 1.81 - SWA 1.09 (2-1)
QPR 1.90 - SOT 0.99 (2-1)
RDG 1.20 - EVE 1.67 (1-2)
WBA 1.53 - CHE 1.34 (1-1)
NOR 1.40 - MUN 1.95 (1-2)
FUL 2.25 - SUN 0.48 (2-0)
WHM 0.98 - STO 0.94 (1-1)
Reply | Threaded
Open this post in threaded view
|

Re: GW 12 Score Predictions

@shots_on_target
Administrator
Alrighty, here are my predictions for GW12.  Entered into the minileague thing.


Reply | Threaded
Open this post in threaded view
|

Re: GW 12 Score Predictions

El Traca
So, another gameweek beckons.....

I have been playing with creating my own model these past few weeks. Basically I've decided to go with a Poisson logarithmic model, pretty much based on the equations proposed in this paper: Analysis_of_sports_data_using_bivariate_Poisson_models.pdf 

However, I find that creating the model based on that is too time consuming, mainly because of the need to make the estimate some of the parameters using full season data and because I've been using SPSS which I think is none too nimble for this kind of thing (I'm learning R so I think if I find the energy I will develop a new one there).

Anyway, I've tried to simplify the equations into a single Poisson log-linear model. I've defined the constant parameters in the equation as the average of the log-mean of certain aspects of the whole league, such as average of goals in an away game for a team and total shots in the box per home game. These constants (which are actually consistently updated as the season goes on  ) define a team offensive and defensive parameters as the deviation from the constants. The formula also includes the interaction parameters describing the difference in the offensive and defensive abilities of the competing teams.

The model is still far from perfect, paradoxically I think that it is both too simple for a start and putting too much emphasis on redundant factors as the constants. For what it's worth, these are the likeliest results according to the model for this week:

Arsenal 2 - 1 Spurs
Liverpool 1 - 0 Wigan
Newcastle 1 - 0 Swans
Man City 3 - 0 Villans
WBA 1 - 2 Chelsea
Reading 1 - 1 Everton
QPR 1 - 1 Soton
Norwich 1 - 3 Man Utd
Fulham 0 - 1 Sunderland
WHU 1 - 1 Stoke