I know ML is used in real horse-racing and other sports, where team/player history matters and can be used as a predictor for future games.

What about for "simulated" games, where the outcome is decided by probability?

Here's an example of what I mean:

Suppose there are 3 tracks T1 to T3, and 12 horses H1 to H12. Every week, all 12 horses are randomly assigned to the 3 tracks (4 horses per track) and assigned betting odds, ranging from 2-to-1 to 15-to-1. The horses in a track are also given a random type of feed (e.g. the horses in T1 might be given hay, while the horses in T2 might be given barley. The feed is random per track.)

As the player, you can place up to 8 bets on all 3 tracks. As the week goes on and other players place bets, the horse odds may increase or lower. At the end of the week, the horses race and you win back some money if your bet(s) hit.

In terms of available data, we know:

  • The starting and current odds of each horse.
  • The win history of each horse (for example, H1 might win 55-65% of the time).
  • The overall strength of each horse.
  • The feed given at each track.
  • The above data for the past 1000 races, including the winners.

Only the odds are known to influence the outcome of the race for sure. It is unknown if the other variables (history, strength, feed) affect the outcome, or by how much.

I ask because, even though past history is given, there's a good chance the winners are simply decided by the current race's odds rather than by history. Since this is a simulation, the outcome is likely decided by randomly generating a number for each track and seeing what range it falls into.

Given the above information, would this something a machine learning algorithm or neural network could predict reasonably accurately? And if so, what would be a good starting point to look at?

  • $\begingroup$ How 'random' is your simulation? Prediction with ML is not much more than applied statistics, so the technique will be more successful if there are useful patterns (However, even with complete randomness, some strategies are better than others, so an ML algo would be better than just choosing something fully at random) $\endgroup$ – Discrete lizard Feb 14 '17 at 8:19
  • $\begingroup$ I feel like this is an area that could be analogous to applying Neural Networks and ML to do stock trading or things along those lines. It might be worthwhile to investigate any research done on that front to see how you can relate it to your horse racing/betting problem. $\endgroup$ – spektr Feb 14 '17 at 18:43
  • $\begingroup$ @Discretelizard It's hard to say. One of the reasons I'm looking at ML algos is to see if there are patterns in the data, i.e. if things like past races affect the current one, or if each race is independent of each other and only dependent on the horse's odds for that race. It's unknown at the moment. Would it be useful to apply an ML algo to this problem to find such patterns? $\endgroup$ – mlnnraces Feb 14 '17 at 19:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.