Predicting NFL Coverages with Bayesian Methods
We created a Bayesian Machine Learning model that classifies plays into 7 coverages (Cover 0-6) based on the pre-snap location of the defense. Our model aggregates 7 one-against-the-rest Bayesian Logstic Regressions with uninformed priors. We wrote a four page report containing details on our methodology and results, which can be accessed here. We also gave a brief presentation on this paper, which can be accessed here.
We used Python and a Bayesian Machine Learning package, pymc, for this project. The code was written in a jupyter notebook, with very detailed comments and notes. Our code, which is a detailed and commented jupyter notebook that is easy to follow, can be found here.
Disclaimer: We used the 2023 Big Data Bowl Data in our creation of this model. This data is not to be used for any other purposes beyond the outlines of this paper. For further details on the competiton rules, access the 2023 Big Data Bowl page on Kaggle.