A Fun Dive into Machine Learning: Predicting NFL Game Winners

05 September 2024 | Kaidan Hetzer

My Machine Learning Model Predicting NFL Game-Winners Takes the Field

The Intern Project

When I was an intern at IG Labs, one of my projects was to create my own machine learning model. Being a huge football fan, I decided to create a model that would make inferences on upcoming NFL games.

My intern days are long past and I’m now a full-fledged member of the IG Labs team, but, my “passion project” has stood the test of time. I’ve continued to maintain my NFL predictor over the past few years and the model has demonstrated an accuracy rate of 65%.

The Challenge

Now, I’m ready to put my model to the real test and go public. Throughout this upcoming NFL season, IG Labs will publish my model’s results each week, despite the potential risk to my reputation and employment (not really, this is all in the spirit of fun!).

The Model Details

The model uses open-source data from pro-football-reference.com and up-to-date betting data from various gambling services for training.

The data and models are updated weekly, ensuring that the most recent games are always considered in their predictions. All methods for data collection, data sanitization, model training, and model inference, are 100% free.

The Predictions

Each week, we will publish the model’s predictions for game winners in the upcoming week. Starting Week 2, we will also address the model’s performance from the previous week. We will also release the model’s Super Bowl predictions.

The Caveats

The model is entirely reliant on team data—incorporating individual player data is too robust to do for free.
The training data is mostly reliant on team stat averages per game (rush yds per game, pass yds per game, etc.).

So, join me this season to track my model’s progress and determine whether I get to keep my job!