[djchapm]

Not sure if you all follow NCAA - but they have a Kaggle competition where they provide you with quite a bit of 'team' level data regarding the season wins/losses and various metrics for each game.  Idea is to see who can develop the most accurate predictions in the NCAA Tournament - 64 teams in single elimination ending in a championship game. 

I had a lot of fun doing this.  It was my first stab at doing something outside education.  I created a classifier for each team by using logistic regression on all their season games.  Then for every match up I'd take the teams' model and plug in the season-average view of the opponent.  Then do that in reverse.  Spits out a probability of win or loss.

I don't know much about college teams or how their season works - but I've made some decent predictions that my in-the-know friends didn't make.  Found where I lost was in the 'big conference' matchups against the 'small conference' matchups.  Apparently... they don't play on the same levels throughout the season - some teams play more difficult schedules, others play easier schedules.  They don't often match up "Duke University" with "Iona" for example.  So - my frame of reference was off a bit for each team. 

It was cool though - found out how much time you must investigate and mold data to your needs, found that to be the more exciting part compared to constructing the network.

DJ