Newsroom

Matthew Littman, MSBA ’20, is one of the proud creators of a new Super Bowl prediction model. Starting from the first Super Bowl in the 1966-1967 season, Littman and his team used seasonal data from NFL.com, pro-football-reference.com and topendsports.com, to decide who will take home the trophy for Super Bowl LIV. Littman’s model has identified the San Francisco 49ers as the Super Bowl champions since he finished his analysis in Week 13 of the current season. He predicts a San Francisco 49ers versus Kansas City Chiefs Super Bowl. 

Littman’s findings were published on the digital platform Towards Data Science. In his most recent analysis, he argues that the most significant data in predicting a Super Bowl winner includes the total defense touchdowns, the offense passing yards per game, the successful offensive 4^th down percentage, defensive sacks and defensive rushing yards per game.

The creation of the program stemmed from a group project in one of Littman’s classes at the UCI Paul Merage School of Business. Inspired by the unpredictable nature of the sport, Littman–a longtime fan of the NFL–took an interest and decided to modify his original idea and develop it outside the scope of the classroom. 

He said, “I have always found it interesting that certain things can’t seem to be predicted, football being one of them, and I wanted to give it a shot and see if you could actually predict a Super Bowl winner, using data.”

“It’s always inspiring to see our business analytics students take their skills outside the classroom to work on projects they are passionate about,” said Bryan Muñoz, associate director of recruitment and admissions. “Predicting a Super Bowl winner is tricky business; this is a very ambitious project.”

The accuracy of the model is centered on predicting a winner based off total team stats, not a player’s individual stats. 

“The amount of data that the NFL captures is staggering. So many stats and resources that are available for use, so I had to decide what I wanted to use in order to create my prediction,” Littman said. “I thought using the team stats would be better than using the player stats because in the end, it is the teams playing each other, not individual players. Certain players do have more effect than others and this would be an interesting angle to look at in the future.” 

Variables are, however, an integral part of the accuracy of the program as well as data imputation methods for missing values. Although the team tried several–including a Linear Regression Model and a K-nearest neighbor algorithm– the Multiple Imputation by Chained Equation (MICE) method was the most successful. 

“I’d say the most challenging part was handling the missing data,” he explained. “If data hasn’t ever been recorded, there isn’t a way to check the actual values, therefore, you need to come up with other ways of figuring out if the values that you are filling in are appropriate. It was a huge process that took much of the time considering there were 89 / 240+ columns that had missing values.”

The experience gained from perfecting a project that spanned over 300 hours is something Littman has thoroughly enjoyed. 

He said, “Once you’ve got a method that will lead to an output, every time you change something in the pipeline, you will get a different output. It is very exciting to see how each method will affect that output. Also, as a fan of the NFL, I know who I would expect to win, and so the more that it looks like what I would expect, the more confident I am in the model.” 

Eventually, Littman hopes to use his model to develop a user-friendly app for all who are interested in one of America’s favorite sports.

However, Littman is not the only student at the Merage School who will be predicting Super Bowl winners this season.

David Savlowitz, CEO & Founder, Competitive Analytics and Professor of Predictive Analytics at the Merage School is hosting a Super Bowl 54 Predictive Analytics Event on campus on Jan. 30. The event attracts Fortune 500 executives as well as Merage School students from the MSBA, MBA, and MIE programs.

“A significant emphasis in our unique class is inspiring students to develop innovative forecast models that can be applied outside the classroom,” said Savlowitz. “And applying predictive analytics to the Super Bowl is not only über-challenging, but ambitious and fun as well.”

Update 2/4/2020:

At the Super Bowl analytics event, Littman and his partner David Ruff changed their prediction after fine-tuning their model. He predicted a score of 20-30, only one point off the actual score of 20-31.

“My new approach was much more robust, and my model was able to predict 42 out of the past 53 Super Bowls.  This is almost 80 % accuracy,” he said.

Littman will try his hand at predicting the Super Bowl again next year and will announce his prediction on Week 13 of the football season.

41 MSBA students competed in the analytics event. Littman and Ruff earned the top spot of the night. Of the 253 predictions collected, MSBA students earned the top three rankings and five of the top eight.

“Interestingly, a majority of our student superforecasters do not have a working knowledge of American football, yet by learning powerful predictive analytics methodologies, embracing innovative forecasting approaches, and leveraging Alteryx, Python, R, and Tableau they produced extraordinarily accurate predictive analytics,” added Savlowitz.