PDA

View Full Version : Ga Tech Computer Makes Sense of Madness


bluepenguin
03-19-2008, 02:10 PM
According to the AJC, "Three systems engineering professors at Georgia Tech have created a computer ranking system, called LRMC, that predicts college basketball rankings more accurately than most, correctly listing 26 of 32 regional champs as one of the top two teams in their region in the past eight NCAA tournaments."

The computer's top 5 picks for this year's NCAA:
1. Kansas
2. Memphis
3. DUKE
4. UCLA
5. unc

gotta love those techies!
Clipsfan
03-19-2008, 02:13 PM
According to the AJC, "Three systems engineering professors at Georgia Tech have created a computer ranking system, called LRMC, that predicts college basketball rankings more accurately than most, correctly listing 26 of 32 regional champs as one of the top two teams in their region in the past eight NCAA tournaments."

The computer's top 5 picks for this year's NCAA:
1. Kansas
2. Memphis
3. DUKE
4. UCLA
5. unc

gotta love those techies!

I'm not going to do the research, but how many more did they choose correctly than the actual selection committee? I mean, how many teams that weren't in the top 2 seeds reached the final 4 over the past 8 years?
Clipsfan
03-19-2008, 02:16 PM
Ok, did the research quickly: 13. But only 6 of those were worse than 3 seeds, so basically the algorithm managed to get those 5 3 seeds right (or get a couple of the 4s instead). I seriously doubt it got George Mason.
UrinalCake
03-19-2008, 02:24 PM
I read about this a couple years ago, it's a pretty neat story, a bunch of nerdy professors who are also sports fans coming up with a solid computer model. I believe they even came up with it around the time GT went to the final game. IIRC, they took statistics over a five-year period, then matched them up with the actual tournament results during that same period in order to quantify what statistics were most strongly correlated with performance.

At the time, I felt their system was flawed because results were applied to a year that was included in the original training data. But this was an old article; presumably they've been able to apply it to subsequent tournaments in see how well it works.