Football Tipping. I built a GA system to predict the week to week outcome of games in the AFL (Aussie Rules Football).
A few years ago I got bored of the standard work football pool, everybody was just going online and taking the picks from some pundit in the press. So, I figured it couldn't be too hard to beat a bunch of broadcast journalism majors, right? My first thought was to take the results from Massey Ratings and then reveal at the end of the season my strategy after winning fame and glory. However, for reasons I've never discovered Massey does not track AFL. The cynic in me believes it is because the outcome of each AFL game has basically become random chance, but my complaints of recent rule changes belong in a different forum.
The system basically considered offensive strength, defensive strength, home field advantage, week to week improvement (or lack thereof) and velocity of changes to each of these. This created a set of polynomial equations for each team over the season. The winner and score for each match for a given date could be computed. The goal was to find the set of coefficients that most closely matched the outcome of all past games and use that set to predict the upcoming weeks game.
In practice, the system would find solutions that accurately predicted over 90% of past game outcomes. It would then successfully pick about 60-80% of games for the upcoming week (that is the week not in the training set).
The result: just above middle of the pack. No major cash prize nor a system that I could use to beat Vegas. It was fun though.
I built everything from scratch, no framework used.
