Machine Learning Model: Predicting Football Play (Run/Pass)

Inspiration

We were inspired by William & Mary Football's historic, record breaking season, and we asked ourselves if it was possible to contribute as students to the sustained success of our football program.

What it does

Our project uses a neural network that takes in inputs about game information before a play is run, and aims to determine whether the play was a run or a pass. We've found that with other data sets, our model is about 75% accurate, which we consider a great outcome, considering most data is fairly ambiguous regarding whether it is likely a run or pass.

How we built it

We used data from Pro Football Focus (All games played by a CAA conference team in the 2022 season), preprocessed the data, and utilized the PyTorch library to build our model.

Challenges we ran into

As our first machine learning project, we found it difficult to get started with some of the data techniques and processes, as well as learning a new library.

Accomplishments that we're proud of

We're proud that we were able to learn something new, learn an unfamiliar library, and build a working model that we feel confident about.

What we learned

We learned data processing and analysis techniques, and we also developed our ability to use an unfamiliar library by researching and paying close attention to documentation.

What's next for Machine Learning Model: Predicting Football Play (Run/Pass)

We believe that this model has many applications, potentially on gameday as a strategic tool, as a way to automate the data collection process and save hundreds of man hours a year for coaching staffs, or as a way to scout opposing teams. The basic framework for the model could also be extended to include/output other data points.