SFU Statistics Students Impress the NFL With Their Moves
In just 12 days, a team of graduate students in SFU’s Department of Statistics and Actuarial Science analyzed player tracking data from roughly 7,000 plays with 34,000 routes to create their winning entry for the National Football League’s Big Data Bowl competition.
The team, consisting of Dani Chu, Matthew Reyers, James Thomson, and Lucas Wu beat finalists in the College division from the University of Pennsylvania, Duke, and Carnegie Mellon universities with their “Routes to Success” entry.
Team member Dani Chu says “We were thrilled that we beat out over 100 submissions just to make it to the finals!” He adds, “When we saw the work of the other finalists, we were blown away. To have our hard work recognized specifically as the winning team was the cherry on top.”
The inaugural event was driven by the NFL’s recognition of the growing value of sports analytics. The competition is seen as a way to provide teams with analytically driven ideas that can be put into action on the playing field.
The NFL, in conjunction with Next Gen Stats, provided entrants with six weeks of player tracking data from the 2017 NFL season. The data captured the real-time locations of every player and the ball on the field every tenth of a second.
The SFU group chose to model play success rate and expected points under various passing route combinations. That meant using their data science skills and football domain knowledge to analyze mountains of data.
Lucas Wu says, “The data we tackled wasn’t even able to be opened in a spreadsheet because of its size!”
Using a machine learning technique called model-based clustering for functional data, the group created a suite of tools to help teams evaluate their playbook and prepare for upcoming opponents.
James Thomson explains, “Using the available data we identified routes and analyzed combinations of routes based on their success rates and big play ability. We identified the route combinations that were consistently strong in both categories.”
Matthew Reyers adds, “We then implemented a model that accounts for the location, direction, and velocity of players’ movements to determine which zones of the field are under their control. This can be used to visualize the effect of route combinations on opening up the field for the target receiver.”
After being selected as finalists, the team travelled to Indianapolis to present their solution to an audience representing 32 NFL teams including NFL media, scouts, analysts, and management.
With some last-minute tweaks to their presentation, including the pronunciation of “route” the American way, the team gave their talk, and were ultimately selected as champions of the college division.
The trophy, which now resides in mentor and SFU professor Tim Swartz’s office, is a hopeful harbinger of future successes for the participants.
Each team member has already lined up prestigious summer internships with Statistics Canada, Terramera, the NBA, and Two Hat Security.
Chu says, “We’re all super excited about these internships and hope that some of the connections we made that day will lead to full-time work later on with an NFL team or affiliate organization.”
The team members are thankful for the support of Big Data Bowl staff Michael Lopez, director of Data & Analytics and Jay Reid, senior director, Football Operations Technology Strategy, as well as their academic supervisors from SFU Tim Swartz, Harsha Perera, and Dave Campbell; and SFU professors, Luke Bornn and Tom Loughin.
Read the paper here: https://operations.nfl.com/media/3670/big-data-bowl-sfu.pdf
Video and more on this story from the NFL on Twitter.
By Diane Mar-Nicolle,
Simon Fraser University