Hi,
Has anyone here ever did something like a financial prediction system, but applied to sports betting? For those who aren't familiar with Betfair, it's just a platform that works as an exchange market where players bet on sports events (mainly), it works very similar to the stock market. I'm developing such a system myself on behalf of a Multi-Agent course I'm attending. I built a system that follows the value of the most bet horse (in horse races) 1 hour prior to the race itself. That way the odd isn't biased from the race itself. Most of the agents in the system don't make use of machine learning at all, they have strict betting profiles (they aren't very successful). Right now I'm developing an agent that uses Random Forests to learn market behaviour. I'm feeding it with different combinations of technical indicators that were computed from the odd variation along time. My problem now is, what do I really want the model to learn? I mean, right now it's learning the variation of a moving average. In other words, in a moment t I compute the variation of odd(t) / odd(t+n) where n is the look ahead window I choose. The issue is that even if the model was fitted the data 100% I still have to implement some strategy on top of my model to make the agent bet only when slope of the variation is greater than X or when the variation has been positive for N periods... Can anyone shed some light on the path I should take?
Thanks.
[link][comment]