Data scientist, physicist, and fantasy football champion

First post

Hello and welcome to The Data Stream!

Team defenses and kickers can be incredibly frustrating.  If you didn't draft one of the top teams then who knows what you'll get.  Even if you did you may be up against a tough team in a given week.  And what about bye weeks?  At any point in a season you can be forced to pick up a team defense or a player for just one week, called "streaming".  It's a nice idea to stream these, but how do you pick?  Sometimes a team has a "good matchup", but how much does that actually affect their score?  Here I plan to use data analysis techniques to help stream fantasy football players.  Hence the name of the website.  The Data Stream: using Data to help you Stream.

Wordplay! Buckle up folks, there's a lot more where that came from.

To start I'm just going to look at team defenses (DEF) and kickers (K).  I'd love to extend it to tight ends (TE) and quarterbacks (QB) at some point, but I also have a full time job, so it'll get done when it gets done.  A weekly prediction will look something like the figure below:

 Figure 1: My predictions from last week.  They were pretty close

Figure 1: My predictions from last week.  They were pretty close

This was my best model from week 9.  The vertical lines are the 50% prediction interval for the team's score and the x-axis value is the mean predicted value for the week.  These numbers aren't fantastic, but I generally do get about half of the teams within the 50% interval.  I score my accuracy based on the FantasyPros algorithm (see the Methodology section), and they only require a ranking and not an actual estimated score. 

My plan is to post a prediction every week and write a few methodology articles here and there.  I'd love to eventually qualify to be scored by FantasyPros, but barring that as long as I use this to crush the other people in my league I'll consider this work a success.

Accuracy