On March 1st, Aaron White (Year 2 Statistical Science), Hassaan Inayatali (Year 3 EngiSci), and Daniel Hocevar (Year 3 Computer Science) won the 2023 NFL Big Data Bowl, one of the largest sports analytics competitions in the world. They competed against eight other finalists (and were one of only two undergraduate teams) who were selected from over 300 submissions and 400 participants. Along with the $20,000 prize, the team’s model also has the possibility to be turned into a product used by the NFL’s Next Gen Stats.
The team learned in January that they’d be moving on to compete as finalists in the competition, which earned them an initial $10,000 prize, and an all-expenses-paid trip to Indianapolis, where the team spent two nights in a hotel filled with NFL executives and other finalists. The Big Data Bowl coincided with the NFL Combine, an annual off-season event hosted in Indianapolis where college football players who are eligible for the upcoming NFL draft perform a series of physical and mental tests in front of scouts, coaches, and general managers from all 32 NFL teams.
The team’s model, which uses data from motion-tracking chips embedded in players’ uniforms, provides real-time information on the pocket of space around the quarterback. Visualizations show how much pressure the quarterback is under, and how long it is likely to last. Unlike the binary statistics used in the NFL to measure pressure, this model provides real-time, granular data on the amount of pressure at every 10th of a second, even if there wasn't a hit, hurry, or sack. The model, which took inspiration from a paper by Javier Fernandez and Luke Bornn that models pitch control and player influence in soccer, applies a bivariate normal distribution to determine each player’s influence on the field based on their velocity, distance and location from the quarterback. The probability density functions produced by the distributions for both defense and offense are added up, and the defensive distribution is divided by the total to get the control that the defense has over the pocket. The values obtained then correspond to the pressure around the quarterback.
White, Inayatali, and Hocevar met through the the University of Toronto Sports Analytics Student Group (UTSPAN), where they’re all on the leadership team. The team is supervised by Professor Timothy Chan (Department of Mechanical and Industrial Engineering). After learning about the competition on Kaggle in September, the team set an outline that had targets to hit every few weeks, and continued to meet every week to work on weekly tasks. “[The team felt like] a nice hybrid, where you have your friendships within the club, but you also have to be professional and maintain that sense of a working environment that allows everyone to kind of thrive and develop their own ideas,” said Inayatali.
Part of what set their model apart was its explainability. The dynamic visualizations are simple and clear, making it easy to intuitively understand the end results, and how they’re shaped. The team made a conscious effort to ensure that there were no black boxes, and that every action they took was easy to grasp even for those who don’t have coding skills, like a coach or an analyst.
“The important thing in sports analytics is that you're not working around a group of data scientists. It's a very diverse community, and some people have been working in the industry for 30 plus years, but they've never they've never coded before in their lives, right? So I think the important thing is being able to communicate it to everyone, ” said Inayatali. “And also just having a fundamental understanding of the application. You can't just develop a model and then not have any intuitive sense of what's happening in the actual sport. You need to understand the sport as well, because you need that context for whatever you're developing.”
This summer Inayatali will be working at an internship with the Chicago Blackhawks, White will be coaching baseball at a summer camp, and Hocevar will be working on applying computer vision to sports analytics. The three grew up playing sports, and continue to play in intramurals alongside their studies. Has studying sports from a technical perspective influenced how they feel on the field?
“When you’re playing, you’re relying on your intuitive understanding of the game, but [studying analytics makes you want to know] the exact mechanics of what's going on and, I wouldn't necessarily say the probabilities, but what the best decisions are given a certain circumstance,” said Inayatali. “It’s not like I understand this game better than everyone else, but it's more that I wish I knew more about what I'm doing so that this way I can perform better. I hope all athletes adapt a little bit of that mindset when it comes to analytics. Like, I hope that's the future for sports, right?”