Holding Our Own
Holding Our Own
There has been something of a content backlog as of late, but fear not it is time to return to looking at data driven betting and more importantly how ‘GoalBot’ has fared in terms of trying to predict the Over 2.5 goals market in the English Premier League.
In terms of data driven football predictions, this has arguably been the most successful of all the projects, although with no financial risk attached to it, it either provides more of a safety net or there is a huge slice of irony attached to it.
Would there be less success if money was put on these recommended bets from week to week, knowing my luck there probably would. Then again, as the saying goes you make your own luck and with that in mind, lets take a look at how I fared a fortnight ago.
HOW DID WE DO?
With six matches clearing the necessary ‘GoalBot’ threshold (five Over 2.5 and one Under) it was a return that was a rare miss in fairness, one that only returned three correct from the half dozen listed above.
The games that did get over the line were:
Burnley vs Leicester
Man City vs Crystal Palace
Arsenal vs Sheffield United
While, although I probably don’t need to spell it out, here are the three that didn’t
Brighton vs Aston Villa
Liverpool vs Manchester United
Watford vs Tottenham
Now of course, just three being correct is not the best return but three out of six, absorbs the losses slightly and that means after testing 48 different matches, 35 of them have come back with the correct Over/Under outcome.
Which in percentage terms means ‘GoalBot’ has slipped to 72.9% in terms of success rate, down from 76.1% the gameweek before. That means the weeks goal outcomes has shaved of 3.2% overall.
That admittedly is not ideal but with the sample size increasing by the week, it is fair to say that we are still holding our own and this is a very respectable way return. Something that suggests that this isn’t a fluke.
However, what if there was an extra element that was required and it is one that is highlighted by Watford vs Tottenham who played out a 0-0 draw at Vicarage Road, but ‘GoalBot’ had it done as Over 2.5 goals.
A QUIRK IN THE SYSTEM
Both these teams have a high probability of being in a game which ends with Over 2.5 goals, but is that more down to them scoring or conceding. Well if you look at this neat bubble map, it is clear to see why
The bigger the bubble, the more clean sheets a team in the Premier League has kept and in the case of Tottenham, they are more akin to conceding goals rather than scoring a boat load at the other end at the pitch.
Which also means that perhaps their percentage probability is somewhat inflated and it is actually the conceding of goals in multiple matches, which has then given the North London outfit such a high percentage.
While although Watford have not necessarily been too bad in terms of clean sheets, they have conceded 36 goals this season and more often than not (especially earlier in the season) they have been apart of an Over 2.5 goals outcome.
Again though that has been more down to a leaky defence rather than attacking prowess and therefore, we may well have seen an example of two over inflated percentages going on to cancel each other out.
HOW DOES THIS EVOLVE?
With the data I have to hand, I will have to build a sub-model and see the goal contributions in each Over 2.5 goal game (and under for that matter). In doing this, I can highlight more examples of inflated probability and therefore add in an extra filter.
The upshot of this might mean that there are less games to bet Over/Under on each week, but in doing so it might also give us a pure accumulator each week. For example, the four absolute best chances of getting either Over/Under 2.5.
I will have to see what the data throws up and then decide what rules/thresholds are required but it certainly will provide food for thought over the next few weeks, as I look to roll out a new sub-model on the other side of this annoying Premier League winter break.
Happy punting and thanks for reading. Dan
If this has grabbed your interest and you would like to discuss/feedback then please feel free to drop me a message at firstname.lastname@example.org. While I am always looking for new football/data projects to work on and if you feel that my skills would be of use, I can be contacted at the same address.