Time To Reflect
If you read my previous article (which annoying looks like an absolute horror show when it comes formatting – not sure why), you will be aware that it was once again time to try my luck in the Eastern European hotbed of Belarus, in attempt for data driven betting excellence.
With this being the third week in which I’ve put my money where my mouth is, with either a mixture of both gut instinct and data driven football predictions, it is fair to say that this has been far from a success.
However, not put off by abysmal failure, it is time to once again see how I fared at the weekend and more importantly, have my two predictive models fared any better than the control measure that was set in place last week.
First lets take a look at the traditional six game form model and how I fared in terms of correct picks:
With the control measure being just two correct in week 2 of this project, the bar was rather low in trying to beat it. While I also did not make any changes to my model, as I wanted to test the difference between rotten luck or rotten quality
THE RESULTS ARE IN
And once again, it makes for incredibly bad reading – as once again ‘BelaBot’ has returned just two correct picks out of eight. Once again a 25% hit rate, and not something you can take to the bank any time soon.
Only BATE Borisov and Isloch came good and it must be said that the former were part of an eight-goal thriller against Smolevichy, as the dominant force in Belarus eventually won 5-3 away from home.
Therefore if we review the first three weeks of operation, the scorecard looks as follows:
Week 1: (Gut Instinct) 3/8
Week 2: (Model Control) 2/8
Week 3: (Proper Test, Model Unchanged) 2/8
So on this small sample of evidence, not only is gut instinct seeminly better than the model in question but the model itself is nowhere near the place I want to be and that means I’m going to have to make some changes this weekend.
CHANGE IS A FOOT
Now if you have read some of the previous articles across other major European leagues such as the English Premier League or Bundesliga, you will be aware that one variant of the model can bit the last six home form vs the las six away form and although that is something I will implement, I cannot until we get to Week 13 of the season.
Therefore, the question is what change can be made and I think one aspect could be add to further wait to teams at home and therefore give them more of an advantage when it comes to picking up a win on their own turf.
Another one could be to factor in league position, but that could be too risky at this stage for two factors:
1) The league has 16 teams, so a little volatile
2) It still a little early and the table might need to settle down (Although to my knowledge it might already settled, what do I know)
While again, it is that lack of knowledge that is causing issues, because say if you were working on the Premier League for example, rightly or wrongly your pre-conceived biases could steer your modelling skills closer to the right direction.
Still though, that was always going to be the danger when working on this competition and all I can see at this early stage, is thank the lord for the talents of BATE Borisov, who have at least spared my complete blushes.
But as I mentioned above, there are not one but two models in play at the moment and with ‘BelaBot’ being an absolute horror show, it is time to see how his leaner, more attractive younger brother also fared.
IT’S NOT ALL BAD
Now for the more eagled eye, you will notice that this model spat out a rather risky strategy and it is one that saw no draws being predicted last weekend. With that said, it certainly wasn’t the worst idea, as it returned a much more respectable four in terms of correct data driven football predictions.
Here is the winners rostrum:
While the no draw strategy was a bit risky in fairness, as three of them returned shared points (something that I may need to think about further down the line)
Now when we talk about control measures, this one could not have been any lower as last week this same unchanged model returned just one correct pick (big shout once again to BATE Borisov), so there was definitely some wiggle room in terms of improvement.
Which means, after a rise in success, it makes sense to leave it unchanged and work out what its natural level of success is, the more we test, then the more that additional one out of eight is nothing more than an anamoly.
Obviously it is too early to call in terms of which model is the best, although at the same time it is also rather obvious. Therefore the action plan for this upcoming weekend, is too have a deep look under the bonnet for ‘BelaBot’ and leave its little brother alone for the time being.
In addition to that, I’m also going to serve up some Bundesliga stats and also look at the amount of home wins that have taken place in Belarus this season and whether that give us more of an indication in the functionality of the model. A busy week ahead it seems.
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.