predicting football using r

28
Predicting Football Using R Martin Eastwood

Upload: martin-eastwood

Post on 15-Jul-2015

3.475 views

Category:

Data & Analytics


3 download

TRANSCRIPT

Page 1: Predicting Football Using R

Predicting Football Using R

Martin Eastwood

Page 2: Predicting Football Using R

£1Billion!bet on 2014 World Cup in UK

According to http://www.theguardian.com/sport/2014/jun/02/cashing-in-world-cup-profit-brazil

Page 3: Predicting Football Using R

70%!of all sports betting world wide !is football

According to http://www.bbc.co.uk/sport/0/football/24354124

Page 4: Predicting Football Using R

Not !everybody !loses to the bookmaker

Page 5: Predicting Football Using R

The odds often reflect the!market rather than the true!probability

According to http://www.bbc.co.uk/sport/0/football/24354124

Page 6: Predicting Football Using R
Page 7: Predicting Football Using R
Page 8: Predicting Football Using R

Load The Data Into R

Page 9: Predicting Football Using R

Average Goals Scored

Page 10: Predicting Football Using R

Average Goals Conceded

Page 11: Predicting Football Using R

Home Advantage

Page 12: Predicting Football Using R

Goal Distribution

Page 13: Predicting Football Using R

Goal Distribution

Page 14: Predicting Football Using R

Creating The Model

!

Xi,j ~ Poisson(ɑiβjƔ)!!

Yi,j ~ Poisson(ɑjβi)

Page 15: Predicting Football Using R

Munge The Data

Page 16: Predicting Football Using R

Fit The Model

Page 17: Predicting Football Using R

Predicting Fixtures

Page 18: Predicting Football Using R

Calculate Probability !Per Goal

Page 19: Predicting Football Using R

Calculate Probability !Of Scores

Page 20: Predicting Football Using R

Win / Loss / Draw

Page 21: Predicting Football Using R

Over / Under

Page 22: Predicting Football Using R

How well does it work?

Page 23: Predicting Football Using R

Poor Accuracy For Low Scores

Page 24: Predicting Football Using R

Poisson model assumes scores are independent of each other!!

Doesn’t account for teams motivation - e.g. playing for a draw

Page 25: Predicting Football Using R

Applying the Dixon & Coles Adjustment

x=y=0: 1-λμρ&x=0, y=1: 1+λρ&x=1, y=0: 1+μρ&x=1, y=1: 1-ρ&otherwise: 1

Page 26: Predicting Football Using R

Man City Vs Man Utd

Man City = 1.88 (53%)&Draw = 4.13 (24%)&Away = 4.45 (23%)

Page 27: Predicting Football Using R

ConclusionsPoisson model with the Dixon & Coles adjustment is simple yet relatively accurate!!!

Other parameters, such as weather, injuries, can be added to improve accuracy!!!

Needs combining with an effective staking strategy though

Page 28: Predicting Football Using R

Questions?