[Journey] Predicting Sports Results with Artificial Intelligence

issorc

Senior Member
Joined
Feb 20, 2016
Messages
826
Reaction score
765
Hey guys,

I'm currently working on a project that I've mainly started for fun and learning and that now maybe turned into a possibility to make some money. As I neglected other projects for this it's actually nearly necessary that I make some money.

Background:
I've been programming for 5 years now and worked on many different projects in a few programming languages. I wouldn't call me an expert though because most of the times I worked on my own projects where I did not care about the best or most efficient code. But as I was working as a software developer for a few month and had courses in algorithms and data structures plus tons of private programming experience I'm no beginner either. I was always fascinated by AI and would love to be able to work in the field of Data Science later so I decided to take the first step. Finished a small course about machine learning (there's still a million things left I need to learn) and started in the field of Deep Learning afterwards. After I learned the basics with some test data provided by a course I was taking I wanted to try it on data I gathered myself. And this is where this journey starts (or started) :)

As I first had to get data to train my Artificial Neural Network (ANN) I wrote a small web scraper in Python. I never used python before but I didn't want to waste my time on a beginner course so I just google in case I get stuck. My scraper runs with BeautifulSoup and gathers data about soccer. Right now I do have a dataset with a bit more than 5000 games, including results and additionial stats. I used ~4000 rows to train the network and 1046 to test the network. After days of optimizing the different layers and variables my best model currently predicts 260 out of 1046 games right. That are 24,85% and I'm sure I can hit 25% sooner or later. And we're talking about correct score prediction here, not the winner or a possible draw.

Yesterday came the day I started to use my predictions for betting. I loaded 15€ onto my betting account and had 16,12€ afterwards. I betted before but yet I lost all my money (except 1,12€ that's why I have 16,12€).

When I was betting before I just placed a bet on which team I thought was more likely to win. I did not care about injuries, stats or other stuff. Just based on my knowledge and interest in the sport. I'll bet everything my tool predicts, no matter how stupid it sounds to me. I trust the predictions of the tool much more than mine.

So I used these 16,12€ as following:
22x single bets with a stake of 0,2€
2x combi bets with a stake of 2x 0,2€ and 1x 0,32€
2x system bets with doubles out of 8 with a stake of 0,2€ and therefore 5,6€ for each system bet

The system bets are combinations of the single bets, so every bet inside a system bet is placed as a single too.

Except one BTTS Single, BTTS combi, combi with BTTS and winner, and O2,5 single all bets are correct score bets.

I already created an excel table yesterday to keep track of my performance in the different categories and types of bets. This will be mainly used to optimize my strategy in later stages (if I don't go broke until then)

Current Results:
I know, this wouldn't be interesting without numbers so let's talk some figures here.

Invested: 16,12€

Yesterday's bets:

Combi - BTTS - Lost - -0,32€
Single - Correct Score - Lost - -0,20€
Single - Correct Score - Won - 1,32€
Single - BTTS - Won - 0,18€
Single - Correct Score - Won - 1,04€
Single - Correct Score - Won - 0,94€

Overall: 2,96€

As all three games that I won yesterday are also in one system bet of doubles I already have 27,80€ safe of that bet. Without the stake of 5,6€ that leaves us at 22,20€ that I'll surely profit from that bet.

So total profit for yesterday actually is more like 25,16€

I got extremely lucky yesterday because results were much better as in my test data so I'm excited to see how this continues. I still have 20 bets open and no matter how they end I already nearly doubled my 16€.

I'll update on each day in case I placed a bet for the day before. Two bets open for tonight so expect the first update tomorrow. Both have high odds and are pretty unlikely to happen imo but we'll see.

In case you find that interesting and want to follow along, I'd appreciate your input and thoughts about this. If you have any questions about AI or deep learning, or even web scraping, don't hesitate to ask, I'm happy to help, although I'm just beginning to dig into that topic. I can only recommend it because It'll play a huge role in the future and it's actually pretty easy to start. A guy on reddit once said that for deep learning you need "days to start, but years to master" and that is as far as I can tell, completely true.

Until the next update, I hope you enjoyed reading, although it's not the classical amazon affiliate or instagram journey :p
 
Mind boggling and mind-blowing simultaneously.
All the best, lemme see numbers like XXX $ per day
 
Do you use python to train the ANN too? Good luck!
 
Brilliant use of software.

Be aware that betting companies will ban you if you are too successful. Their businesses are out to take money from you, not to let you make money.
 
Very interesting
Mind sharing your tips here for us to follow?
I'll just drop here what I'm in for today but I'm not sure if I'll share it everyday.
Heidenheim - Darmstadt (2. Bundesliga) 0:0
Lazio - Neapel (Serie A) 1:0

I actually wouldn't advise anyone to bet on this as Darmstadt, Neapel and Lazion have started really good this season and imo are really likely to score more goals.
Brilliant use of software.

Be aware that betting companies will ban you if you are too successful. Their businesses are out to take money from you, not to let you make money.
Thanks! Right now I'm not concerned about this as I still need to figure out how to make constant profit. With correct score tips it just needs one weekend of weird result to let me lose all bets :D
 
Thank you, I'll try to achieve that :p

Thanks!

Right now everything is written in python :) Thank you!

Great journey, i will follow this :)

would be cool if we could be a part of this if you dont mind, ofcourse on our own risk ;)

a daily email or something!
 
Hello I've been thinking doing this stuff with AI for a long time and I find your jurney very interesting.
Good luck or better say Right predictions.
Wich AI course do you suggest? I have experience in Java, algorithms and data structures.
 
You could share the tips here and people SHOULD paper trade for say 4- 6 weeks, to see what the outcomes are.

G/L with it though.
 
Great journey, i will follow this :)

would be cool if we could be a part of this if you dont mind, ofcourse on our own risk ;)

a daily email or something!
Email is out just for the reason that I don't want to take this out of the forum. As some people here seem to be quite interested in the results I'll post predicted results from now on when I'll post my stats for previous games. And thanks for following! :)
Hello I've been thinking doing this stuff with AI for a long time and I find your jurney very interesting.
Good luck or better say Right predictions.
Wich AI course do you suggest? I have experience in Java, algorithms and data structures.
Thanks!
I liked this course https://www.udemy.com/deeplearning/
Their code is in python and R but you won't get around it to learn one of them anyway.
Although I only read a few pages so far I'd recommend the book "Deep Learning" by Ian Goodfellow, Yoshua Bengio and Aaron Courville. It's really comprehensive and it includes the math and everything you need to get a deep knowledge of the topic.
Edit: You can actually read it for free here: http://www.deeplearningbook.org
You could share the tips here and people SHOULD paper trade for say 4- 6 weeks, to see what the outcomes are.

G/L with it though.
Thanks for the input, I'm going to share the tips from now. And paper trading is a really good idea. I really cannot advice anyone to follow my tips at the current stage with real money. And if someone should do, he/she should only use small amounts of money that one really can afford to lose. Over large samples the best I was able to achieve so far was 30% when fine tuning parameters, correct score bets yesterday were 75% right so my profits will definitely be lower as more bets are to come.
And thanks for wishing me luck :)
 
Based on the predictions I placed another two bets for today:

1. Bundesliga - Hertha BSC 2:1 Bayer Leverkusen
Serie A - Cagliari 3:1 Sassuolo

Please note, I would never bet some of the games I'm currently betting, but it is what the program is predicting. My main focus was to remove feelings, emotions and other stuff out of the process and make predictions based on data and stats only. Maybe in a few weeks I'll change how I place bets but currently I'm just doing what my tool is telling me to do.
 
Everything looks strange to me. But not the numbers :P
Although I don't know how your tool will be able to completely predict 'exact scores' of matches. If you keep doubling your money, am sure BHW won't mind joining you.
 
Fascinating.. Hoping you could share your experience in future that can become an inspiration to others.
 
Everything looks strange to me. But not the numbers :p
Although I don't know how your tool will be able to completely predict 'exact scores' of matches. If you keep doubling your money, am sure BHW won't mind joining you.
I'm sure I got really lucky yesterday, but as long as I can beat the probabilities that are included in the odds of the bookmakers I'm good to go :) My next step will be to increase the test sample and add features to the prediction, but that will have to wait a few days or maybe even weeks. As I said, out of the 1046 test results I do have I predicted 256 right. Based on assumed average odds of 7.5 for exact score games I'd have made a 83,5% profit on my "invested" money. Assuming that I'd have betted the same amount on each of the games :) We'll see how good it will perform in the long run, there will be some down phases ofc. I'll keep the units I bet (currently 0.2€) as low as possible to prevent going broke.
Fascinating.. Hoping you could share your experience in future that can become an inspiration to others.
Thanks! :) I'll try to share as much as possible as I move on and work on improving everything. Maybe even someone who reads this gets inspired and finds a way to use AI for other sides of Internet marketing :)
 
Very interesting project! I know zilch about AI, so does it like learn as it bets, and change its strategy? btw have you checked out OpenAI gym?
 
Work smarter not harder! Seems like an amazing idea, good luck. Will be following.
 
I'm sure I got really lucky yesterday, but as long as I can beat the probabilities that are included in the odds of the bookmakers I'm good to go :) My next step will be to increase the test sample and add features to the prediction, but that will have to wait a few days or maybe even weeks. As I said, out of the 1046 test results I do have I predicted 256 right. Based on assumed average odds of 7.5 for exact score games I'd have made a 83,5% profit on my "invested" money. Assuming that I'd have betted the same amount on each of the games :) We'll see how good it will perform in the long run, there will be some down phases ofc. I'll keep the units I bet (currently 0.2€) as low as possible to prevent going broke.

Thanks! :) I'll try to share as much as possible as I move on and work on improving everything. Maybe even someone who reads this gets inspired and finds a way to use AI for other sides of Internet marketing :)

If you get 7.5 in return per game you would had to have 134 games correct to make even.

Let's say you instead was playing over/under 2.5 goals and that the rerurn is 2.0 per game, then you would had to have 523 games correct, would this be more profitable for your merhod?
 
Back
Top