Introducing Pi Premier League
Working at a Business Intelligence and Analytics company has many perks. Yes, it gives me the chance to use the word ‘algorithm’ as an excuse and improve my table tennis skills but most importantly provides me with some tools to help predict football matches.
Karl Marx may have said that religion is the ‘opiate of the masses’ but for me personally it’s football. Bramall Lane is my church, Chris Wilder is my vicar, Billy Sharp is my God and If you haven’t guessed it, Sheffield United are the bible from which I pray.
Now that my beloved team are back in the Premier League, I politely asked my technology hero Ken whether we could use our Analytics engine and my previous work on football predictions to see how we do versus actual humans in predicting the weekly matches.
So, a bit of background… I have taken data from the last 20 seasons and created metrics against this to see if there are characteristics that can improve the chances of predicting an outcome of a football match.
I then ran this through PI Analytics to test the objectives and see which was the most statistically significant and provide something for us to apply to each game. In the coming weeks, I will explain some of the theory and calculations behind the scenes that got us here but for now…
- I had a rating factor (changing on a rolling basis) for each team that then can be applied to each upcoming fixture.
- The difference between the two team factors (Rating Difference) for each game will then predict whether this is a home win, away win or a draw.
As an aside and for some context, statistically over the past 20 seasons of the Premier League the following chance of result has occurred:
Home Wins 47%
- Away Win: 28%
- Draw: 25%
In our analytics model, home wins for example are looking like the below. As you can see the starting node is a 47% chance of a home win but hopefully using our rating we can get this to over 64% chance of predicting a home win.
Written by Alex
Alex is one of our consultants, specialising in helping new partners get up and running with our products and helping existing partners with new installations and the like. He has over six years’ experience in MI and database development.