Close But No Cigar
Close But No Cigar
If you read one of my previous articles, you will be aware that I was once again ready to wake my good friend ‘PremBot’ from his period of festive hibernation and I was ready to use a data driven approach to predicting English Premier League results.
Now admittedly this series of modelling has not been a barnstormer it must be said and with a success rate of less than 45% in terms of correct individual outcomes, you would be in a similar or betting position by just picking by gut feel alone.
One excuse is that the Premier League has been very volatile this season, although in honesty it does seem as if things have started to settle down as of late and therefore, we cannot rely on that particular reason anymore.
Anyway enough of me trying to get my excuses in, let’s now take a recap at the picks that my good friend ‘PremBot’ unleashed over the course of the weekend and whether or not I am now on the road to riches.
THE MOMENT OF TRUTH
One key thing that you need for a 10-fold accumulator is for your bet to still be live by the time the Saturday 3pm kick-offs come around, because if an earlier game stiffs you, then the rest of it is nothing more than a fruitless exercise.
Thankfully after events at both Bramall Lane and Selhurst Park the 382/1 10-fold was still very much on, the question then was whether or not the other eight games in question could help me get over the line.
Unfortunately that was not the case, although to be the honest that was always going to be an incredibly ask. However, with seven of the ten ‘PremBot’ picks being correct, it was the models closest week yet.
The previous best in a week was six out of ten, so there was one step better this time and around and weirdly with no changes under the bonnet of model to my last attempt at this, when I recorded a paltry three correct bets out of two.
IS IT BROKE, DOES IT NEED A FIX
After the weekend of returning just three out of ten, it would have been very easy to rip it apart and start again. However, due to a combination of laziness and time constraints, I decided to leave it alone and it did pay a relevant dividend.
That said, it is still not perfect and you get the feeling that some work still needs to be done and not just for me but for West Ham and Arsenal, as any ‘new manager bounce’ was not in effect over the course of the weekend.
That in itself is interesting beause I did also do some A/B testing where all the picks above were the same bar swapping out wins for both the Hammers and the Gunners, in doing so running two parallel 10-fold.
In hindsight, I should not have bothered, as neither outfit won at the weekend and therefore I should have had full confidence in my own predictive model, still there’s no harm in trying a slightly different approach.
Talking of an approach, this data driven betting one has now tested 80 games and 39 have been predicted correctly by ‘StatBot’ – not great reading it must be said, and that works out at a 48.7% hit rate.
A level that the man off the street could probably return himself, so ‘PremBot’ is yet to show his overall worth to me at the moment. Then again, performance such as this, does show that the time and energy is not a complete bust.
It may be something of a dream but to be three games away from a 382/1 longshot, is not the worst effort and if this can be the norm in terms of performance, then who knows what might be around the corner.
Anyway ever the eternal optimist, I have also built a new variation of this model and it is one that now combines the final result prediction and the Over 2.5 goals prediction, which means ‘PremBot’ and ‘GoalBot’ have merged into ‘ComboBot’, more on that development later in the week.
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.