Data scientist, physicist, and fantasy football champion

Week 10 DEF predictions

Hello fellow data enthusiasts!

For the last few weeks I've been analyzing Fantasy Football data (and even used it one week!), but I wanted to score myself against FantasyPros experts.  They said I needed a website, so... here we go!  My next few posts will probably have a bit more intro, but since it's already Thursday let's just get this week on the books.

 

For the last few weeks I've tried out a few different models, and the ones that seem to work the best are simple linear models.  I'm going to give them letters for now to track them across weeks.  Write in the comments if you have a better idea.

I don't know how to add a comment section yet.  Or if I even want to.  I probably don't.  Y'all are trolls.

 

Model A:

Model A-1.png

 

This is the model that uses most of the parameters I have right now. It uses:

  • The team that’s playing
  • Who their opponent is
  • Their predicted score
  • Their opponent’s predicted score
  • Whether the team is home or away

It also specifically only uses the data from 2016. I’ve tried models that include both 2015 and 2016 data but they seem to be less accurate. This was the best model from weeks 8 and 9.

 

 

Model B:

This is the model that uses most of the parameters I have right now. It uses:

  • The team that’s playing
  • Who their opponent is
  • Their opponent’s predicted score

Like model A it also only uses the data from 2016. This was the 2nd best model from weeks 8 and 9. It uses fewer variables. A simpler model for simpler people. Really, I was using it to avoid overfitting, but it doesn’t seem to matter too much. Maybe on a longer time scale it would work out better?

Model C:

This has the same parameters as Model A:

  • The team that’s playing
  • Who their opponent is
  • Their predicted score
  • Their opponent’s predicted score
  • Whether the team is home or away

But it uses the data from 2015 and 2016. It doesn’t do quite as well as model A, but I want to start it now to get an idea of the accuracy. I’ll likely have to use it for the first 4 or 5 weeks next year.

Week 10 model summary

RankModel.AModel.BModel.C

1LA SDLA

2KCLAGB

3SDKCKC

4MINPHIARI

5GBMINHOU

6PHIGBBAL

7ARIARICAR

8CARBALPHI

9BALNYJMIN

10CHICARWAS

11NYJMIANO

12MIACHINYG

13WASWASSD

14DENDENDEN

15ATLATLNE

16SFSFCHI

17NONOMIA

18HOUTENPIT

19TENHOUATL

20NYGNYGNYJ

21CLECLEJAC

22JACJACSEA

23SEATBTB

24NENETEN

25CINCINCLE

26TBSEASF

27PITDALDAL

28DALPITCIN

Models A and B give very similar results. There are a few small differences here and there but overall there aren’t enough differences to really matter. Model C is quite different in a few ways. I’ve been thinking about picking up SD this week, but I already have LA so it’s probably notnecessary.

 

Week 10 K predictions