05/08/24
March 18, 2009
Computer Model Likes North Carolina to Win Tournament: Georgia Tech professor Joel Sokol is getting some press for his NCAA Tournament computer model, which successfully predicted last year's final four, the championship game and the winner. His model, NCAA Basketball Ratings using a Logistic Regression/Markov Chain (LRMC), predicts top seeds Louisville, North Carolina and Pittsburgh and second seed Memphis as the final four and North Carolina defeating Memphis for the crown. The biggest first-round upsets in the model are No. 12 Arizona over No. 5 Utah in the midwest and No. 12 Wisconsin over No. 5 Xavier in the east.
posted by rcade to basketball at 12:40 PM - 11 comments
Hmmm ..
The Computer model seems to have Duke and UCLA rated higher than Louisville (Louisville is the 7 seed, Duke 5 and UCLA 6); Gonzaga, Kansas and Missouri rated higher than Oklahoma ....
Someone had better check that machine for malware before it infects the entire university network !
posted by cixelsyd at 01:12 PM on March 18, 2009
posted by BoKnows at 01:51 PM on March 18, 2009
Well lookie here, I have the Tar Heels winning it all in my bracket too. Sweet!!
posted by BornIcon at 02:20 PM on March 18, 2009
which successfully predicted last year's final four
I'd be interested to see how he does this year. Last year was kind of a funny one with so many higher seeds winning. Not to say his model doesn't work, it would just be more impressive if it picked some major upsets across a few years.
posted by yerfatma at 05:15 PM on March 18, 2009
From the article
Sports analysts made a big deal of the loss, he said, which seemed statistically dumb to Sokol. As far as he's concerned a close game might as well be a tie -- the toss of a coin could predict the winner.
"That last shot is almost meaningless," he said.
Uh, yeah, OK. I think I'll stick to my stream of consciousness, pick em in a daze while my son plays in the bath method.
And guess what, I had UNC as well.
posted by cjets at 05:48 PM on March 18, 2009
Not to say his model doesn't work, it would just be more impressive if it picked some major upsets across a few years.
I doubt that kind of thing is actually possible. Assuming that the group picking the seeds are rational (contentious much?), they're using more or less the same data that the computer is. A major 16 over 1 upset would likely require the seeding committee to overvalue the 1 seed or undervalue the 16 seed in a major category.
Also, given the nature of injury and its effect on the outcome of teams, it's really hard to predict what will actually happen over time. It's neat that it worked perfectly once, though.
posted by dfleming at 08:40 PM on March 18, 2009
Huh, I have North Carolina over Memphis in my bracket. This is some nice news.
posted by Ying Yang Mafia at 11:26 PM on March 18, 2009
How do you account for Lawson's toe in the model?
posted by bperk at 11:18 AM on March 19, 2009
From the article:
"Somebody said that we're not in President Obama's Final Four, and as much as I respect what he's doing, really, the economy is something that he should focus on, probably more than the brackets." - Coach K
I say fuck that. What does it take, 5 minutes, maybe 10, to fill out a bracket? Yes he's the president, but he can still be a sports fan, right?
posted by BoKnows at 07:10 PM on March 19, 2009
I doubt that kind of thing is actually possible. Assuming that the group picking the seeds are rational (contentious much?), they're using more or less the same data that the computer is.
If that's the case, there's little point to the exercise. To clarify, I'm not asking for 16 over 1, I'm asking for 12/ 5 upsets, correct calls of 8/9 games, that sort of thing. Last year wasn't the best test for a prediction engine.
posted by yerfatma at 02:50 PM on March 20, 2009
Shouldn't there also be a computer model/equation that would predict the effect predictions have on the teams in question?
So, NC would have won, if there had been no model that predicted them to win in the first place.