Friday, May 6 2022
11:00am - 12:30pm
Masters Presentation
NFL Play Call Prediction Using Sequential Neural Networks

The prevalence of data analytics in professional sports has significantly
increased over the last 20 years. First popularized in
Moneyball: The Art of Winning an Unfair Game (2003), the
use of advanced analytics is now mainstream in the four major
U.S. sports and abroad. In the National Football League
(NFL), millions of dollars are invested into analytics departments
and data is being used to drive decision making at every
level of an organizationís operation. These departments can
leverage statistical methods to learn the oppositionís tendencies,
providing a substantial competitive advantage. In particular,
the defensive team can improve its strategy by accurately
predicting the offensive teamís play call (whether the play is
a run or a pass). To this end, many prior works have implemented
machine learning algorithms for play call prediction.
However, none of the works encountered have treated play by play
data as sequential. In Football, the offensive teamís current
play call is dependent upon the sequence of plays called
before, therefore, there is a time series component that a modeling
strategy must account for. In this work, we explore the
ability of sequential deep learning models to predict NFL play
calls. Namely, we compare the performance of Recurrent Neural
Networks (RNNs) and Long Short Term Memory (LSTM)
networks to baseline models (Logistic Regression and Gradient
Boosted Decision Trees). Using classification accuracy and
ROC-AUC as metrics, we found that sequential models outperform
the baseline.
Speaker:Joseph Director
Location:Zoom link in email

Download as iCalendar