NFL Big Data Bowl

February 25, 2019

The NFL Big Data Bowl was a competition for assessing player tracking data and identifying optimal receiver-route combinations. Graduate students Dani Chu, Matthew Reyers, Lucas Wu, and James Thomson, under the mentoring of Professor Tim Swartz, have been selected as finalists to present alongside student teams from Duke University, the University of Pennsylvania, and Carnegie Mellon University to NFL analysts, front office officials, and members of the media at the 2019 Scouting Combine in Indianapolis.

Their specific take on the problem of receiver-route combination assessment involved the use of functional clustering to build their own version of a route-tree with the data provided. From there combinations of routes on a given play could be better analyzed in terms of their effectiveness to create either positive results or high probability plays.  

Positive results were viewed through the lens of plays that generated positive expected points, a metric common to the NFL analytics community. The use of this metric allowed the group to identify route combinations that could serve any coaching need from short reliable plays to long and reasonably consistent plays.

High probability plays were explored through the idea of space created by the offense. This work involved updating accessible models, such as Voronoi Spaces, with speed and direction measures. Borrowing from the work of Dr. Luke Bornn, the group brought a more accurate perspective to the idea of field ownership for both defense and offense.

The work ultimately looks to encourage a further push for analytics in the NFL. By focusing not just on the task of building optimal receiver-route combinations, the group has essentially forged a toolkit for analysts including tools for route identification, route clustering, play review, and space ownership. Further, their report details how the tools are used, examples of usage, and conclusions of interest to NFL analysts. Their coming presentation at the NFL Scouting Combine seeks to demonstrate the usefulness of these tools and to lay the foundation for future SFU collaboration.