Pitt prof's prediction machine picks Atlanta to prevail in Super Bowl
Kostantinos Pelechrinis isn't aiming to inherit the mantle of the late Jimmy the Greek, but he just might.
Although Las Vegas odds makers give the New England Patriots an edge in Super Bowl LI, Pelechrinis, an associate professor of computer science at the University of Pittsburgh, is betting the Atlanta Falcons walk away with the Lombardi Trophy.
Vegas sportsbooks have made money on 24 of the past 26 Super Bowls, but Pelechrinis has data on his side — seven years' worth of NFL statistics to be specific.
“It will be a pretty close game, but Atlanta has a small advantage,” said Pelechrinis, who leads the Network Data Science Lab in Pitt's School of Information Sciences. “In the case of this game, they have a 54 percent probability of winning with a very good standard of certainty.”
Pelechrinis and Carnegie Mellon University alumnus Evangelos Papalexakis, a professor of computer science at the University of California, Riverside, partnered for a year on a study that examined statistics from 1,869 regular and postseason NFL games to build a probability model for their Football Prediction Matchup (FPM) engine.
Papalexakis' research focuses primarily on using analytics to predict what word the brain will focus on when you read something on a screen. But he enjoys sports analytics. Such studies, he said, have a way of making complex computer science widely accessible.
“This is a great way to showcase your message. You can attract a lot of students with examples like this that are easy to understand,” Papalexakis said.
Pelechrinis said he's seen the same interests among students at Pitt.
The researchers' study, published in the journal PLOS, analyzed various game factors — including turnovers, penalty yardage and the balance between passing and rushing — that correlated with the probability of a team's victory.
Pelechrinis then ran thousands of simulations of a Patriots-Falcons contest through FPM to come up with the probability of a Falcons win.
While some might beg to differ with the computer scientists' conclusions, Pelechrinis said FPM already has proved its worth in the NFL playoffs.
Pelechrinis said it gave the Steelers a 72 percent probability of winning against Miami on Jan. 8, a game which Pittsburgh won 30-12. FPM gave the Steelers a 50 percent probability of winning against the Kansas City Chiefs, which Pittsburgh narrowly accomplished, 18-16, on Jan. 15.
And finally, unlike many hopeful Steelers fans, FPM accurately predicted an AFC Championship loss to the Patriots — giving Pittsburgh only a 36 percent probability of prevailing Jan. 22. New England won the game, 36-17.
Papalexakis marvels at how well the FPM engine has performed.
“This is working as well as a proprietary system that someone paid a lot of money for, and we did it with data that anyone can access,” he said.
Debra Erdley is a Tribune-Review staff writer. Reach her at 412-320-7996.
