Data scientist, physicist, and fantasy football champion

Week 1 DEF Predictions

Welcome back dear reader(s?) to another wonderful year of fantasy football! It’s great to be back. After a long summer of searching for new models, the search is finally over. I’ve learned some new modeling techniques to make better models and I’ve got some new ideas to make my life easier (shoutout to R Notebooks!), so I’m expanding into QBs this year as well. Keep an eye open for those soming soon.

For defenses, my plan for this season is to keep only the best models from last year and to make a few new models that I think have a chance of really nailing it from week to week. If I’m keeping a model from last year I’ll keep its name but update the years. For example, last year Model A used only the data from 2016. This year it will use only the data from 2017. Take a look below for a table of the DEF models I’m using this year.

I’m no longer going to fit the worst models from last year. Even my worst DEF model did alright, but it’s a new season and it’s time to take out the garbage and focus on the future. See you in hell, Models B, D, and E!

And for those of you excited about the rookie model for this year, check out my article from a few weeks ago about my hot new Bayesian model, Model G. Model G is a Gibbs sampler version of Model F, which means it should give roughly the same rankings, but the prediction intervals (i.e., my confidence in the rank) will be slightly different. It did great in my simulation of last season, so I’m excited to try it out this year.

Here’s an overview table of all the models this year:

ModelTypeScoreOppScoreOppHomeAwayDataYearsPointSystem
ALMXXXX2017IOUH/Standard
CLMXXXX2015-2017IOUH/Standard
FLMXXXX2015-2017TruScor
GGibbs SamplerXXXX2015-2017TruScor


The models are pretty homogenous this year. They all include the same terms but with different data, point systems, and methods of statistical inference. These models consistently beat the other ones that included other terms, so until I come up with new data to predict with this is about it for now.

On to the predictions:

 

Model A:

Coming soon (when I actually have enough 2017 data)

Model A parameters:

The team that’s playing
Who their opponent is
Their predicted score
Their opponent’s predicted score
Whether the team is home or away
Model A Notes:

Uses only data from 2017
Standard points
Did well in 2016, but didn’t stand out


Model C:

.Model C-1.png

Model C parameters:

The team that’s playing
Who their opponent is
Their predicted score
Their opponent’s predicted score
Whether the team is home or away
Model C Notes:

Uses data from 2015-2017
Standard scoring
Did well in 2016, but didn’t stand out

 

Model F:

.Model F-1.png


Model F parameters:

The team that’s playing
Who their opponent is
Their predicted score
Their opponent’s predicted score
Whether the team is home or away
Model F Notes:

Uses data from 2015-2017
Truscor ignores defense TDs but weights interceptions and fumbles to try to account for it
I really liked this one in 2016. It was marginally better than A and C
 Model G parameters:

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

Model G:

.unnamed-chunk-2-1.png


Model G Notes:

Uses data from 2015-2017
Truscor ignores defense TDs but weights interceptions and fumbles to try to account for it
Uses Bayesian statistics (Gibbs sampler) to create a model
Would have done very well in 2016 based on simulation

Week 1 model summary

RankACFG
1NADENPITPIT
2NAPITDENDEN
3NAHOUCARCAR
4NABUFHOUHOU
5NAINDBUFBUF
6NALACLACLAC
7NACARDETDET
8NADETGBIND
9NAOAKINDGB
10NAPHITBWAS
11NAMINWASTB
12NATBOAKOAK
13NAGBLARNE
14NALARCINCIN
15NAATLNEATL
16NAWASMINMIN
17NAKCPHILAR
18NANYGMIAMIA
19NAMIABALBAL
20NABALATLPHI
21NAJACARIARI
22NANESFDAL
23NAARIJACJAC
24NASFNYGSF
25NACINDALNYG
26NASEANYJNYJ
27NADALKCNO
28NANYJNOSEA
29NANOSEAKC
30NACLECLETEN
31NATENTENCLE
32NACHICHICHI

 

Conclusions


PIT, CAR, HOU, BUF, DEN, LAC. Those are the teams to play this week. Those are all good teams expected to win lower scoring games against teams that tend to give up a lot of fantasy points. No real thoughts here. If you have one of them, play them.

A note of caution against PIT: they’re rated so highly this week becuase they’re expected to trounce CLE. This is based on last year’s data and therefore last year’s CLE QB debacle where they consistently gave up huge points to opposing defenses. I don’t really know what to expect from CLE this year, but last year my motto was “never bet against betting against Cleveland”, and I think that still stands until we see otherwise.

GB, IND, and DET are also good plays, but a little riskier. Again, these are good defenses, but going against teams that aren’t necessarily known for giving up a lot of points. Opponents are expected to score less than 24 points, though, so they’re still solid plays.

Rounding out the middle are MIN, TB, CIN, and WAS. MIN had a good defense last year, but putting a defense against NO is always risky. TB did well last year, but they’re playing MIA this week, and I don’t know what to think of the return of Jay Cutler. CIN against BAL is also pretty heavily QB-dependent, so we’ll see who they play. Depending on which Carson Wentz shows up, WAS might be in for a tough day.

Finally, a quick note about LAR. This week they’re going against Indy, which is generally not recommended (see: Luck, Andrew: touchdown rate). However, Luck is out this week and while I don’t have any way to model whether the DST is going up against the starter or a backup, I did include the change in predicted scores. That wasn’t enough to push them even into the top 12 this week, but if you’re picking between one of those last few defenses this week they’re probably worth a start since you can usually expect another few sacks and interceptions with a backup QB.

I wouldn’t be excited about playing anyone else, but if I had to pick a model and guess I would trust Model G or F. We’ll see over the next few weeks which of these models is the best. I’ll also try and compare myself to other pros, likely Yahoo since I can get their predictions easily.

Good luck everyone, and to the people in my league: good luck vying for 2nd place.

Week 1 QB Predictions

Model accuracy throughout the year - Kickers