Football/Soccer/Premier League Prediction. Curiosity made me learn Data mining.

shuga101shuga101 Member Posts: 1 Contributor I
edited November 2018 in Help
Hello everyone

Great forum.
Been poring over the sections related to my subject.

Foremost, I will like to say I am more adept at using Weka....I have never tried RM but the discussions on football prediction here made me join the forum.
Even at that, my expertise in Weka is limited to the purpose that aroused my curiosity. ..that said...i confess my weka and predictive analytic skills are still poor but my 'imagination runs wild'.

Now to the matter at hand.
Over the past few weeks I have been observing the predicted results of data I spent many months cleaning,  the dataset has 12 attributes with the last one being a class value for the results...Home. Draw. Away.

I run a bayes classifier on it...it is the most successful for the purpose so I have never bothered with other ones with the exception of a few whose names are localized for weka I presume.

The results for each weekend prediction is:
20-40% accuracy for Draws
20-50% accuracy for Away wins
30-70% accuracy for Home wins
The fluctuation is as a result of updating the data with new ( current weekend soccer games) dataset with empty class values for it to predict. This i took as trained and test data.

The accuracy above is not the weka software prediction accuracy but the real life accuracy/results of the matches played.  What this means is that for this upcoming weekend 29th November...if I input the matches to be played and run the bayes (Aodesr and bayesnet) algorithm. ..I will get a set of predictions and when the matches all play out later in the day on tv...If there are 15 draws as shown on tv....weka would predict between 2-5 or more of those matches correctly....the bad thing is that the weka algorithm would predict maybe 20 draws for me and never pinpoint which one in particular would successfully end in a draw....same for home and away wins.

So I am here on this forum to brainstorm with intelligent folks on how to find out how a predictive analytic software can pinpoint/narrow down to 1 or 2 or more predictions from the many others it has predicted. E.g...if weka predicts for me 12 draws...I can "stake my life" on it that at least 3 of the predictions will turn out as predicted but I wouldn't 'bet my fart' on which one in particular will be correct.

So I devised a plan which has won me games/money in 3 straight strikes.  It is still based on luck but my decision is based solely on the data presented by weka.

Last week November 21 2015, weka predicted 11 draws...14 away wins and 17 home wins...my bet plan was to pick 8 matches accumlator from the away and home wins and bet them as "any team to win"...the odds were low ....so I paired the 8 matches with one match from the 11 draws weka had predicted....meaning I placed a bet on 9 games accumlator...(1 match to end in a draw and 8 matches that any team should win)
I did this eleven times....(remember weka predicted 11 matches to end in a draw)
So from the eleven draws, I selected one draw match and paired it with the 8 matches...then I repeated the process till I used up all the eleven draws....

3 matches ended up as draws and and no team played a draw amongst the 8 teams I paired along wit the draws... I made 900% profit on my investment..I have my receipts to prove....I tried this method last weekend and one team messed up the whole plan....out of last week weka's prediction of 18 draws...(there was 5 draws for real as shown on tv)..but one team out of the 8 teams I bet "not to end in draws" actually ended in a draw...the sad thing is that one of the other algorithms saw it as a draw but I disregarded it.

So the reason I am here is to ask
(please note I am a newbie with newbie imagination) if there is a way to visualise the tree/journey for each prediction that "actually" ends up accurate so that it can be compared for correlation or studied for patterns or rules or associations that will help increase accuracy against the next set of predictions....

I particularly would like to hear from Sebastian and Martin as it was their discussions that encouraged me to join the forum.


Sign In or Register to comment.