[Journey] Predicting Sports Results with Artificial Intelligence

[fredfredricks]

You should start selling some tips to us hard core gamblers

 

[Marcel Freiberg]

Hi there,
i am really impressed by the OPs Idea of predict scoccer games with the help of KI. Actual I see some good things like no prejudice, or favors: in fact the most rational bet you can place.
On the other hand it is kind of hard to predict a game by the historie of matchdata like it is impossible to know the next lottery numbers by looking at the past.
So does your KI get additional Information like the health of the players, does it weight the most recent games higher then old ones?
I wish you the best of luck with this journey because it is one of the most amazing here

 

[SEO Sniper]

Good luck OP

 

[issorc]

At first, this weekend the predictions were really bad. Sorry for everyone who might have placed bets on this. I won 3/34 (roughly 9%) of the bets with odds of 12,7 and 9 which in total adds up to 28 won units against 34 placed units (one unit is 0.2€ for me right now). That makes up to a loss of 6 units. With some weird system bets (I nearly all lost) this weekend brought me a total loss of 23€. Nevertheless, I'll go on (most likely only with single bets) and continue to work on this. Even if it does not earn me serious amounts of money I really enjoy working on it. I started to dig deeper into the topic but digging through all that math in a language that is not native for me is super hard. Luckily, I basically know all the concepts through my studies at the university. I have to take the math course again so maybe that will even gain me an advantage for that

A few days ago a user made me aware of a super cool project he's been launching at Kickstarter. @Chiefjop and a few other guys created a software that enables you to backtest betting strategies over different years and leagues and compare the results to find a strategy that (hopefully) works. After that you get notified when a bet occurs that match your strategy. The kickstarter page completely gained my interest and this project is something I'd really enjoy doing or using myself so I donated for the project at kickstarter myself. If you are interested to backtest strategies over huge datasets and see immediate results this tool seems to be a great choice. The banner in my signature leads you directly to the Kickstarter campaign I guess that @Chiefjop will be happily answer any questions regarding this

are you using machine learning?

As imo deep learning (and therefore neural networks) are a sub category of machine learning I'd say that yes, this is based on machine learning.

I read on a roulette forum some person recorded over a million spins over 7 years, also made similair thing but aimed roulette (live wheels with a dealer). It could predict for 70% which number the ball would fall. Don't know if its real but if it is, that would be awesome!

So OP I hope you would achieve that also one day

Honestly I doubt that someone could achieve that with Roulette. I've never played it myself but isn't it completely based on randomness?

This has HUGE POTENTIAL! you could retire in few years, but be aware you could make some enemies down the road

Thanks mate, there's definitely huge potential in it but to uncover this potential it will need much more data + better algorithms. We'll see how far I can improve it

You're right, Pinnacle has the correct score market on very few matches and it's a shame the others you've checked don't allow German bettors.
Maybe just one of them might allow Germans and that will be enough - their odds aren't that different from Pinnie's.

Just remembered an old news that Betfair has withdrawn from German market too years ago, not sure if it's still the same.

Still, there are hundreds of bookies, many of them reputable, you can find lists all over the net, one for example here:
https://www.bookmakersreview.com/ratings/all


On topic about the prediction method itself. Someone already posted about the odds you're getting need to be considered.
Absolutely.

The thing with sports betting is this: you can't win in long term unless you are betting on value, i.e the odds you bet on (converted in probabilities) are higher than the true probability of the score to actually occur. This probability is intrinsically there, it exists, though no one knows its exact distribution because it depends on so many factors. All we can do is try and estimate it as correctly as we can.

So if we only bet on odds with value, then we will turn profit in the long run, no matter what - the Law of large numbers dictates it. For this to happen though we need:
-------------------------------- ----------------------------------------
1. To correctly estimate the true probability with as little margin of error as possible.

2. Find odds that are higher than that.
--------------------------------------------------------------------------

On the other hand, if we only bet on the most likely outcome, being it predicted by algorithm or handicapper, and disregard the odds we are getting - just bet on whatever odds the bookie has to offer, that won't cut it. And that's even if we assume the prediction for this outcome to be the most likely of all possible outcomes is absolutely spot on.
If we are consistently getting odds lower than the true probability of the outcome to occur, then we will lose money, no matter what.

Is it possible that your algorithm is finding value even without you considering it actively? Well, it's not impossible, but seems unlikely, given that it doesn't even consider the odds for its prediction. For it to find value consistently this way it has to be that the bookies are consistently underestimating the probability of the most likely (according to your predictor) outcome for some reason and offering higher than fair odds.

Thanks for the list of bookies A mate recently started betting on Betfair so I guess it's available in Germany again.

And I really appreciate your input regarding the value stuff. The problem right now is that my tool just gives me the result it guesses. No information about the probability. This is why I'm right now testing different results to see which results it is the best at predicting to convert the probabilities of these results to the odds. Over long term I'll have to change this and adjust the tool but right now I'm playing around with this. I'll change that once I got more knowledge about statistic distributions and machine learning/deep learning/AI. It's a long way to go for me and the chances of succeeding are quite small but that is fine I use this project to improve my skills in all these areas and play around with some money Even if it won't be able to provide good enough bets long term it might be helpful in case I should be looking for a job in the data science area later

Very interesting journey mate! I will definitely follow this to see where you are at in a few months. Do you think/know if there are other people doing this? I could imagine that its similar to the trading markets where big whales are using bots on a large scale. I am not too familiar with sports betting but I heard if one is fast enough the biggest gains could be made using live bets as soon as something unpredicted happens which in my opinion could be done with bots too.
Anyway please keep us updated and good luck on your journey mate.

I heard about a startup in Great Britain that does analyse player's position by the tv pictures and maps it onto a 2d map and analyze how likely a situation is to end in a goal by data it has gathered before. This is a really great stage of in-play analysis There are for sure some betting millionaires out there. Take a look at the owner of Midtjylland (Denish soccer club). This guy made a fortune with sports betting and now runs several soccer clubs and tries to improve their game by data analysis. As far as I know they even won the Denish Championship after they were bought by that data analysis sports betting guy Thanks for the wishes though and I'll keep you updated

I'm seeing some teams from Portugal, nice

Yes I'm currently analyzing all major European Leagues when running my script Sporting against Porto was one of the few bets I won last weekend.

Hahaha me too, It sounds very interesting. Good luck OP!!

Thank you!

You should start selling some tips to us hard core gamblers

These tips are anything but not worth to be sold On some games it would've been better yet to roll a dice

Hi there,
i am really impressed by the OPs Idea of predict scoccer games with the help of KI. Actual I see some good things like no prejudice, or favors: in fact the most rational bet you can place.
On the other hand it is kind of hard to predict a game by the historie of matchdata like it is impossible to know the next lottery numbers by looking at the past.
So does your KI get additional Information like the health of the players, does it weight the most recent games higher then old ones?
I wish you the best of luck with this journey because it is one of the most amazing here

I guess previous match data plays a role in predicting a future soccer match while previous lottery numbers are worthless but I get your point I'm currently trying to increase the amount and quality of data for the network but it's really hard to get enough historical data about everything I'd like to add to train the network. The most recent games should be weighted more by the network because it should learn over time that the correlation between the most recent result to the output is closer than for the other results Thanks for the wishes

Good luck OP

Thank you

 

[midnightlee]

At first, this weekend the predictions were really bad. Sorry for everyone who might have placed bets on this. I won 3/34 (roughly 9%) of the bets with odds of 12,7 and 9 which in total adds up to 28 won units against 34 placed units (one unit is 0.2€ for me right now). That makes up to a loss of 6 units. With some weird system bets (I nearly all lost) this weekend brought me a total loss of 23€. Nevertheless, I'll go on (most likely only with single bets) and continue to work on this. Even if it does not earn me serious amounts of money I really enjoy working on it. I started to dig deeper into the topic but digging through all that math in a language that is not native for me is super hard. Luckily, I basically know all the concepts through my studies at the university. I have to take the math course again so maybe that will even gain me an advantage for that

A few days ago a user made me aware of a super cool project he's been launching at Kickstarter. @Chiefjop and a few other guys created a software that enables you to backtest betting strategies over different years and leagues and compare the results to find a strategy that (hopefully) works. After that you get notified when a bet occurs that match your strategy. The kickstarter page completely gained my interest and this project is something I'd really enjoy doing or using myself so I donated for the project at kickstarter myself. If you are interested to backtest strategies over huge datasets and see immediate results this tool seems to be a great choice. The banner in my signature leads you directly to the Kickstarter campaign I guess that @Chiefjop will be happily answer any questions regarding this

I'm very interested in your journey, good luck!
I see the project in kickastarter, it's very promising! Donated hope to see a thread by @Chiefjop

 

[cacats]

Hey! Any predictions for this weekend?

 

[kingdombunniez]

I'm interested in this as well

 

[Universe0]

 

[issorc]

Hey! Any predictions for this weekend?

I'm interested in this as well

Sorry guys, nothing for this weekend. I'm currently heading home from a vacation and had no opportunity to predict the games. Next weekend I'll post predictions again

 

[sumitup88]

can check 10bet out as well .... they do take punters from Germany

 

[cacats]

Hi! How are you doing? Any updates?

 

[issorc]

Hi! How are you doing? Any updates?

Thanks for asking, I'm really sorry but I'm not going to bet on this weekend. I'm currently working on a completely new system which i'm setting up in cooperation with a friend. I hope to be able to use it next week but maybe it'll take a bit more time. Really sorry for everyone who expected new predictions this weekend.

 

[issorc]

Hey guys, I do have some good news I worked hard this week on a new system that includes a database of more than 105k games over the last ten years and various leagues. I wrote a few python scripts that work on the database and do some statistical analysis (provided by my friend) to provide me with odds for BTTS and Over 2.5. This way I can compare my odds to the odds of bookmakers to find the ones that offer value.

Things that are on my to-do list right now and will be done in the next weeks or days:

• Automate all scripts and everything to spit out my odds every weekend automatically
• Create a small webinterface to output everything in a human friendly way
• Get odds to find value bets automatically without looking at the bookmakers
• Get as many historical odds as possible to backtest my algorithms
• Apply machine learning algorithms to improve the analysis
• Watch out for ways to gather more (relevant) data

I still like the idea of correct score bets so I'll implement that sooner or later too again. Especially since I saw today that sime sites do offer historical data for correct score bets which would help me a lot in terms of backtesting.

I'll post my predictions for saturday and sunday in this thread so everybody can follow again Nothing for tomorrow as the program does not have them finished yet and I won't be at home tomorrow to post them, really sorry!

 

[deathass01]

sounds great @issorc! Keep up the good work!

 

[midnightlee]

@issorc keep up! i'm very interested

 

[cacats]

Sounds great... Keep Up, mate!

 

[irishengteacher]

Hi issorc.

Fortunately/unfortunately I have plenty of experience with gambling and trying to figure out systems.

Regarding getting banned you can always use Betfair Exchange and you will never be banned.

I bought a soccer stats crunching piece of software a while back, it doesn't predict anything just helps you sift through data and identify value. I found it cool, it is someone's hobby project that keeps getting bigger and bigger. Perhaps it or its databases might be of use to you.

It's here: http://www.cgmbet.com/

 

[AmRyanThomas]

Hard work Beats Tallent, if Tallent didnt work hard

 

[Xsoid]

Hey, for everyone who is interested in this topic here, the search engine bing is not bad in predicting the results of an game.
You can read more about it here: qz(dot)com/233830/world-cup-germany-argentina-predictions-microsoft/

Hey guys, I do have some good news I worked hard this week on a new system that includes a database of more than 105k games over the last ten years and various leagues. I wrote a few python scripts that work on the database and do some statistical analysis (provided by my friend) to provide me with odds for BTTS and Over 2.5. This way I can compare my odds to the odds of bookmakers to find the ones that offer value.

Things that are on my to-do list right now and will be done in the next weeks or days:

• Automate all scripts and everything to spit out my odds every weekend automatically
• Create a small webinterface to output everything in a human friendly way
• Get odds to find value bets automatically without looking at the bookmakers
• Get as many historical odds as possible to backtest my algorithms
• Apply machine learning algorithms to improve the analysis
• Watch out for ways to gather more (relevant) data

I still like the idea of correct score bets so I'll implement that sooner or later too again. Especially since I saw today that sime sites do offer historical data for correct score bets which would help me a lot in terms of backtesting.

I'll post my predictions for saturday and sunday in this thread so everybody can follow again Nothing for tomorrow as the program does not have them finished yet and I won't be at home tomorrow to post them, really sorry!

@issorc You said that you have more than 105k games in your database.
According to qz(dot)com/259556/after-crushing-the-world-cup-microsoft-is-predicting-the-nfls-whole-season/ they wrote "A full 256-game NFL season will be tough to predict accurately. While the performance in the World Cup’s knockout rounds was impressive, the model was less successful during the more frenetic and much larger group stage of the tournament." And that is right. I don't know how many leagues you are using at the moment, but keep in mind that more is not always better. You can cover up a wide range, but you need a massive amount of data to cover up the range and also get usefull results.
I don't remember if you use other variables than the game database, but you can take a lot other variables with in your database. To be honest I am doing the same like you. I also make an semi ai to predict some results and love ai, machine learning, etc.. The last weeks I talked sometimes with my father about an ai which can predict all soccer results perfectly. Yes that can never become true, but we talked about what variables influence an soccer game. There a lot of useful things you can use:

• Player statistics (google sofifa and then you can find nice stats from the fifa about the player)
• The weather (was/ is it sunny, rainy, hot, warm, ... all can influence the performance of the team a lot)
• You can use whoscored(dot)com to find many many useful informations about the games
• Search how was the percentage of the fanbases from the teams in the station (don't think that there is an online source for that, but it's worth a try)
• You can also use Twitter, Instagram, etc.. I can't find the paper at the moment, but they described in there how the predicted the outcome of an game only using Twitter. That's very interesting in my opinion .
  
• ...

I hope I can help you with these proposals.
Good luck man, you can do it

 

[issorc]

Thanks for all the input again, I'll answer later guys, just dropping my predictions for now

I only cared about games for today yet because I don't have any money left right now and do have to wait for the bets to finish

Silkeborg - Sonderjyske U2.5 @1.7
Pescara - Avellino O2.5 @1.47
Khabarovsk - Ufa U2.5 @1.3
Lokeren - Charleroi U2.5 @1.85
Sligo Rovers - Derry City U2.5 @1.85
Sandecja Nowy Sacz - Cracovia Krakau U2.5 @1.8
Vitoria Setubal - Maritimo U2.5 @1.44
Kerkyra - Panetolikos U2.5 @1.4
Coventry - Colchester U2.5 @1.95
Coventry - Colchester (Colchester U1.5) @1.28
Blackpool - Wigan (Wigan U1.5) @1.8
Rennes - Lille U2.5 @1.61
Levante - Getafe U2.5 @1.57
Cesena - Foggia O2.5 @1.85
Sheffield United - Reading (Reading U1.5) @1.33
Fc Zurich - Grasshoppers Zurich (Grasshoppers U1.5) @1.36
Huddersfield - Manchester United (Huddersfield U1.5) @1.12

In total I spent 19,29€ so I'll update tomorrow about my return