Picking Brackets with the Experts: Bracketology Science

Josh Duke

March Madness is upon us. Even if you are not a fan of basketball, chances are you might still end up filling out a tournament bracket. After all, the art of filling out an NCAA Tournament bracket has become a cultural moment—and a big business. Just doing a quick search online or browsing through newsstands will quickly reveal the amount of time and energy people put into filling out their March Madness bracket, with outlets ranging from ESPN to The Washington Post giving advice on how to make the perfect bracket. We have even coined a new word: bracketology (which, by the way, does not mean filling out any old bracket, but rather the NCAA Tournament bracket).

With so much attention spent on making that perfect March Madness bracket, researchers like James Du have taken notice. Du, assistant professor of Sport Management at Florida State, has an interest in a number of subjects, but recently he has been working on how to incorporate machine learning techniques to determine sport brand loyalty. However, with the beginning of March Madness fast approaching, Du realized that machine learning techniques can be applied to bracket building.

“A good example would be to use machine learning techniques to select an optimal set of measures and metrics by taking advantage of archival data published at NCAA.com,” Du explains. “This would allow you to project head-to-head results when picking a winning bracket for March Madness.”

Rise of the Machines

Machine learning is a term that you might have heard a few times at this point. Essentially, it is a branch of artificial intelligence research that looks at automating analytical model building. In other words, machine learning is about teaching machines to draw conclusions from and identify patterns in data. Machine learning appears in a number of ways, but a few examples include Netflix recommendations, self-driving cars and even financial fraud detection.

Du believes that machine learning can assist in designing the perfect bracket, specifically a type of machine learning called swarm intelligence. Using an algorithm, swarm intelligence evaluates the picks of thousands and thousands of March Madness would-be bracketologists. What makes this approach so interesting, says Du, is the fact that the system relies on “bad” picks.

“Counter-intuitively, the [swarm intelligence] model is driven by picks from a pool of the least informed sport fans rather than so-called ‘experts’ to make a prediction,” Du says. The problem with these experts is that they’re often “biased by overthinking sleeper picks, as by definition they will have to share ‘expert opinions’ that are not similar to each other.”

Whereas artificial intelligence research tries to replicate human thinking, swarm intelligence takes the opinions of people to extrapolate patterns, which ultimately approximates human thinking. Researchers utilized swarm intelligence in the past to pick winners of tournaments, movie awards, and more—and with surprising accuracy as well.

March Madness Meets Machine

“Although artificial intelligence and big data can vastly optimize the prediction accuracy, a perfect projection is still unrealistic given that the odds against predicting all 64 games correctly are 1 in 9.2 quintillion,” says Du. “With that, using a mixed strategy (combining machine learning models and your gut instincts) would be wise to put down on your bracket.” By combining both your fan picks and machine learning insights, Du says that you will be able to “focus on an overall picture rather than the results of individual match ups you have created.”

While Du cannot say that machine learning will give you a perfect bracket every time, he believes that the application of such advanced analytics will only get better over time. Whether or not this year’s machine-designed bracket turns out to be perfect or a bust doesn’t matter as much to Du as the data he’ll gain from this year’s tournament.

If you’re interested in learning more about sport analytics, cutting edge technology, and implementing machine learning in the sport industry, click here to learn more about the Sport Management program at Florida State.