Thread: March Madness
View Single Post
Old 04-02-2013, 11:37 AM
hemphill hemphill is offline
Join Date: Jan 2013
Posts: 18
Default Re: March Madness

I used logistic (also probit) regression on NFL football this last season. This was before I knew anything about machine learning. I used just win/loss records between the teams, and used the entire part-season record to generate a fit, then predicted the weekly outcomes. I compared my results with an NFL commentator who also predicted weekly outcomes. My results in the beginning of the season were terrible, since the method didn't have any prior information to go on. About two-thirds of the way into the season, it was performing better than the commentator. Then it performed poorly in the post-season, for a total record that was slightly worse than the commentator. I attribute the post-season results to the fact that there was additional information about the teams that was not reflected in the win/loss totals for the season.
Reply With Quote