Follow Datanami:
March 1, 2018

It’s March, Time For Bracket Analytics

Here’s an optimistic prediction for picking the perfect bracket in the NCAA men’s basketball tournament, aka, “March Madness”: The odds have risen to as “high” as 1 in 128 billion, according to a DePaul University professor cited by the NCAA as it beats the drum to fill out office brackets.

Professor Jeffrey Bergen presents the upside; the downside is that other mathematicians put the chances of a perfect bracket (correctly picking 63 winners, one of which eventually loses to the national champion in a 64-team field) as low as 1 in 9.2 quintillion.

To narrow those all-but-impossible odds, the NCAA is again working with Google Cloud and its Kaggle data science community and competition platform to help prognosticators spot that elusive first round upset that busts most everyone’s bracket. At least one other analytics vendor, Adobe Systems (NASDAQ: ADBE), is also offering free access to its enterprise-grade platform in the run up to the tournament.

The futile pursuit of the perfect March Madness bracket is again being used as a data science teaching moment by Google-owned Kaggle. The March Madness competition also provides a vehicle for combining the search giant’s cloud, BigQuery and Cloud Datalab platforms to crunch at least a decade of college basketball stats to narrow probabilities as long [and wide] as Professor Bergen’s white board. (shown below)

DePaul University math professor Jeffrey Bergen and his “optimistic” odds of a perfect bracket prediction.

The overarching idea behind the competitive application of predictive analytics is to augment gut instincts with a dispassionate, data-driven approach to picking winners and losers.

(Full disclosure: This reporter is a University of Wisconsin fan. In last year’s second round, only a few basketball analysts gave the Badgers much of a chance against NCAA tournament top seed and defending national champion Villanova University. Those analysts noted that the fleet, athletic Wildcats often fared poorly against grind-it-out, half-court teams like Wisconsin. Lo and behold, the Cheeseheads upset heavily favored Villanova, 65-62, blowing up many brackets.)

To help spot this year’s upsets, Google and Kaggle said they would provide starter code and Google Cloud credits along with tutorials on how to apply machine learning techniques to developing a winning, albeit imperfect, NCAA basketball bracket.

The bracket analyst coming closest to perfection wins $25,000, with $15,000 for second and $10,000 for third place. Total prize money for NCAA Machine Learning Competition is $100,000.

The NCAA began using Google’s analytical tools last year to help determine often contentious tournament selections and seeding. Variables such as strength of schedule are used to determine, for example, which schools make the lucrative Division 1 basketball tournament.

Meanwhile, other analytics vendors are also getting into the bracket racket. Under the rubric of “metrics” madness,” Adobe said this week it is unleashing its enterprise analytics data tool to bracket-ologists. Adobe said it has loaded college basketball data into its analytics platform and will make it available during the tournament to help raise the extremely long odds of spotting the upsets that ruin most brackets.

Public access to the Adobe Analytics “Hack the Bracket” is here.

Recent items:

Deep Learning is About to Revolutionize Sports Analytics. Here’s How

Google Cloud Will Assist With ‘March Madness’ Brackets

–Editor’s note: The story has been updated.