by Jason Salas
Guam - While preparing for the first Saturday of college football for the 2011 season, I took a dinner break from studying charts, rosters, matchups and the pre-season Top 25 polls to watch one of my favorite shows, Modern Marvels on History Channel. It just so happened that the episode I caught profiled the technology used to manage crops.
And remarkably, in a vignette about how grapes are picked en masse, the producers strongly emphasized a major advantage used in such work to achieve optimal systemic results: the harmonious synergy of man and machine.
About 20 minutes hence, the episode having concluded and a delicious berry salad already beginning to digest in my belly, my thoughts returned to all things pigskin, wondering if the Bowl Championship Series, which utilizes the average of two human polls and a computer poll, might actually generate more reliable, reproducible results if it were run exclusively on machine learning. Let me break it down for you.
The major source of contention in the BCS era has always been the unfair differences derived between the inexplicable insight of human intuition against indisputable mathematics. To date, computers with static intelligence haven't been able to comprehend to implied merits of a team valiantly mounting a comeback, hitting a game-winning field goal at the end of a contest, earning a 7-6 victory; we likewise can't standardize a way for human pollsters to uniformly appreciate a Top 10 team getting blown out by an unranked, out-of-conference, FCS team at home in the opening week of the season.
(Cough, cough, Michigan.)
This is the bane of the BCS's existence…and has been since its implementation. The noble pursuit of having the true #1 and #2 teams meet for the title to avoid the problems of the bowl system has generally worked, but is wrought with controversy over who truly is deserving of the top spots. I say we take our appeal to a higher court.
Instead of testing ad hoc methods that are sketchy at best and relying on equally imperfect guesstimation, let's allow learned resources to structure a system that interprets all polling data from 1998 to 2010, look for patterns in predefined major categories, and develop an ability to accurately predict the country's best teams. As grape-picker systems show us, we can marry careful subjective design with automated objective execution, so no one can complain about overfitting in either direction and creating data spoilage.
We've got thirteen years of information on which such a program can chew, changelogs documenting the various modifications and alterations to formulas like strength of schedule and margin of victory, and a history of wonky math deducing multidimensional inputs like home wins versus victories on the road.
Such a system would be wisely informed, staunchly unaffected by tradition, peer pressure, or the lure of financial gains (and losses) by corporate partners and/or conference affiliation.
Moreover, our NCAA overlords might consider in earnest employing data mining techniques as a means of pinpointing instances of that most cruel and unusual punishment - human bias. Similar work is seen in biology circles to identify anomalies to detect disease. Which, considering the application, would be poetically sublime.
I mean, really. Consider the base argument: innovation has allowed us to develop efficient, effective systems to pick grapes…why not college football's national champion?
Jason Salas is KUAM's resident BCS nerd and writes his column "Poll Positions" every Saturday. Follow college football all season long with CBS on KUAM-TV11 and Notre Dame on NBC on KUAM-TV8.