Using Graph Theory to Predict NCAA
Interesting...
http://blog.biophysengr.net/2012/03/eigenbracket-2012-using-graph-theory-to.html
He has us at #60 and losing to Ohio in the first round.
The only real beef I have with his system is that he values even games too even and does not factor in home-field advantage.
But this should be pretty interesting to follow. It seems like EVERYONE is picking Belmont to go far.
Wish he would use it on past tournaments and see how he succeeds then. I would think that it's extremely hard to find consistent success with any formula when it comes to the tournament.
Overall, I think it's a great idea but it could be refined after testing decades of games to get a higher success rate, or one that completely destroys the idea. The #1 and #2 seeds should always be on top though in almost any formula or approach though, so that doesn't necessarily mean it's a good predictor of success, just that it's not terrible. I better predictor would be getting consistent success in the closer-seeded games (8 vs 9, 7 vs 10, etc.)
PageRank (which he did not use) would in theory be a much better indicator for this type of problem than eigenvector centrality (which he did use).
It is a mix of completely conventional choices (three #1-seeds and one #2-seed in the Final Four with Kentucky winning -- duh) with completely hopeless upsets (Harvard!? and Belmont in the Sweet Sixteen). Harvard beats Wisconsin? I at least have Harvard winning in the first round, and like the Eigenbracket, I have Ohio University beating Michigan. I hope I am wrong, but I have a bad feeling about all of this "Ohio" nonsense our Athletic Department has been laying down. Karma. It's a bitch.
Oh and down goes EigenFinalFourPrediction Syracuse...
I actually used pagerank a few years ago to pick my bracket. It did well, especially in the first round or so, but only ended up getting 3rd in my bracket pool.
His top 10 doesn't have Duke, who beat 4 of the top 10 teams. Ok.