I've always wanted to enter the NFL #BigDataBowl and 2021 was the year I finally got the skills to be able to enter.
For those that don't know what the #BigDataBowl is, I'll sum it up here:
Kaggle competition
NFL provides access to the public for their coordinate data
Judges include NFL Team Staffers and NFLHQ Staff
Finalists commonly get recruited by Sports Teams for their analytics staff
$25k to the Winner
This year's focus was on Special Teams (Kicks or Punting plays). Not much analytics has been done on these plays outside of Justin Tucker and his ability and no public metrics focusing on gunners/vices etc. (specialist positions).
Given this is a competition, I entered to win. To win a public competition where the previous winners are some of the smartest people in the world enter, you need to stand out. That's where my ideas came from where I had three main ideas:
I had three ideas:
PlayerTV
Special Teams Usage
Punter's Impact on Kick Returner Yardage
Special Team Usage was to create a 4th down model, 2 point conversion model and an onside kick model. 4th Down and 2pt conversion models are common and there are lots of models around. I don't believe there is an onside kick model around. This idea would have taken into account of a lot of different factors but mainly focused on:
Personnel
Formation
Players
Punter's Impact on Kick Returner Yardage is an interesting approach to the whole Yards over Expectation models. I presumed a lot of focus would be on blockers and the returner but not the punter. A lot of the factors impact returning yards, but I wanted to look at three main ideas:
Expected Bounce Yardage (Where will the ball bounce)
Expected Bounce movement (after bounce yardage)
Expected Returner Decision (Return, Fair Catch, Bounce)
The idea that I went for was PlayerTV. No one has done something like this so far, but it makes so much sense. Lots of people focus on the metrics (pure numbers), and therefore standing out from the crowd is hard. I didn't have the data that I wanted to do the Special Teams Usage metric, and I thought that a lot of people would look at Return Yards over Expected (which they did). Therefore neither of my metric ideas made sense to do if I wanted to win.
PlayerTV comes from the idea that public-facing camera angles aren't good enough at showing the player's point of view (POV). This idea would help both the staff and the public understand what each player sees and why they make their decisions.
I noted down everything related to my PlayerTV idea in the Kaggle submission, but I thought it would be best to go through all my shortlist ideas I came up with.
Link to Kaggle Submission: https://www.kaggle.com/zacrogersuk/playertv-by-zacrogers-excluding-appendix
Link to Twitter Thread: