Future Timeline

So this seems easy, you can make a machine learning model fast and with no prior knowledge.
We need some good training data, from the article, he had:

"The horse's ID
The dates of the horse's previous races
The track where those races took place
The distance of each race
Whether the race was run on a turf or dirt surface
The horse's finishing position in each race
The purse, or money that the horses were racing for
The implied probability, a proxy for the betting odds given on the horse at the start of the race
The trainer's ID
The jockeys ID
The number of the times the horse has raced
The age of the horse"

I know nothing about horses, but maybe I could make some model for soccer. The question is, what data and where to get it.

https://spectrum.ieee.org/betting-with-no-code-ai

It's not that simple. My friend wanted to use machine learning to win big at sports betting. I pointed out that the house probably uses machine learning to set the odds so his model would have to do better than theirs to win money. And they probably have smart people making those models.

Of course, it's not simple.
Here's a guy who has looked into making a ML model. He did some research but decided not to do it.
First of all, we don't have to write code, if we use the No-code AI, so it's easier that way.
Secondly, he stopped partly because the Bookmakers limit the winning gamblers. I know of one bookmaker that does not limit: Pinnacle sports. There may be others. They actually welcome good players, because they can adjust the odds accordingly.

The article gives some useful links. We learn that they put the odds themselves into the ML model.
https://towardsdatascience.com/my-findi ... c8684baa8c

I have a clear plan now. I found a research paper where they selected 28 features from every match, which they used to train the NNM.
I will target the english premier league, find the data and build the NNM, using the No-Code AI. Maybe I will use the 20/21 season data and the start of this season. Then I can update the model every week with new data.

If nothing else, it's a good practical project for setting up a NNM.

So I signed up for a 14 days free trial at akkio.com. I am impressed with how easy it is to set up at machine learning model.

I found some data for the English premier league 20/21 season. (CSV file).

I trained the model with the following features:

-Full time result ( Home, Draw, Away)
-Home-Team
-Away-Team
-Home team shots total
-Home team shots on goal
-Home team shots off goal
-Home team possession time
-Home team corners
-Home team fouls
-Home team yellow cards
-Home team red cards

And also corresponding features for the away team.

This is a preliminary model, just to see if it works. Also, these data were freely available. I might need to do some data extraction from websites to get more detailed data, or I could pay for it.

So this is just 9 features (excluding team names). This paper uses 44 features which is later reduced to 28:
"Improvement of Football Match Score Prediction by Selecting
Effective Features for Italy Serie A League " MANAS Journal of Engineering

Then I pulled odds from Pinnacle Sports for upcoming matches, converted to probabilities and then compared to my model:

For example,

Newcastle-Leeds Pinnacle prob.: 32.3 27.8 41.7 My model: 61.7 13.5 24.87
Burnley-Arsenal Pinnacle prob: 25 28.6 47.6 My model: 26.6 31.2 42.3

In all I have 8 matches, where Pinnacle and my model agrees on the result in 6 of those matches.

So this is a proof of concept, with only 9 features from the 20/21 season (no data from the current season), the model seems to function quite well.
There's a webinterface to the model, so you can see for yourself, at least for the next 14 days:
https://app.akkio.com/deployments/x8LXNZDmsgbx9aocm1Cg
Choose home team and away team and then click on predict fields.

Next I will try to find some more quality data to add more features. Also I must add data for the current season.
A problem is, how many seasons should one include in the model? Also, last season was different with no spectators.

When set up, it's a really powerful prediction model. It can predict any feature that you want.

There are different training modes when making the machine learning model at akkio.com.
The default is 10 s mode, which I used for the model above. Because of some error, I couldn't see the button where you can adjust the training mode. When I zoomed out in the browser ( Chrome) the button eventually showed up.

I trained a model for 5 min, which should give better quality. There's also a 30 min training mode available.
When comparing the 10 s model with the 5 min model, there was a huge difference. Now the probabilities are closer to the bookmaker's probabilities, which you would expect to be very close to the "true" probabilities. This is very uplifting and tells me it's going to work.

I'm working on finding some high-quality data for the complete model. It's not easy to find data which are free. I could webscrape but that will take me a while to figure out.

Obviously, I'm continuously exploring my options and learning.
Are there other places where you can do no-code AI? Yes, a lot, here's a top 18 list: https://serokell.io/blog/top-no-code-platforms

I might try the Google Cloud Auto ML. It would be interesting to compare the results across ML models.

I looked into Google AutoML. What a mess.
The current platform akkio.com is so simple to use, but of course it's not free for much longer.
I still haven't found more data for the training.

There's a guy on Twitter, Vladimir Haltakov, that is putting together a machine learning course. The focus is on practical applications. I will try to follow the course.

A bit more progress. I have got my hands on some more in-game features of Premier league 20/21, and I have made a 5 min rendering ML model. I believe it's too short of a rendering, the website has a 30 min model, but it costs money.

It would be very interesting to work with a ML model, try different features, test it on historical data. It takes a lot of time and I need a place where compute is cheap. I also need to grab data from different websites and put it in csv files. The course I mentioned by Vladimir, actually starts with how to grab data for a ML model, so that's perfect.

If anyone wants data to play with, just let me know

It's not that simple. My friend wanted to use machine learning to win big at sports betting. I pointed out that the house probably uses machine learning to set the odds so his model would have to do better than theirs to win money. And they probably have smart people making those models.

Actually, this is not completely true. Sure, the bookmakers have large complex models, or they buy odds from someone who does. But the bookmakers have to do something that is to our advantage; balance books. They have to change the odds depending on how the incoming bets are distributed. That could give us an edge if we have a really good prediction engine.