Data scientist, physicist, and fantasy football champion

Week 3 DEF Predictions

This is why I started this whole shebang: streaming fantasy defenses. I wanted to not give a second thought to which team I plugged in each week. Instead, here I am overthinking this. But this is the classic problem when you try to automate something. As always, XKCD has a comic about this. But, that’s life, I guess. You have to spend money to make money, break eggs to make omelettes, and waste hours of your life trying to model fantasy defenses to crush the dummies in your fantasy league. I think Plato said that.

Last week Yahoo and Model C both did well, though there’s not a ton of daylight between any of the models since they’re all so similar. But they’re all so similar because they’re better than the rest. Maybe I should make a new, bad model just to see how it fails?

I finally have the data to start up Model A. Let’s see who we predict will do well in week 3:

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

 

Model A:

.Model-A.png

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

.Model G-1.png


 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 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 3 model summary

RankACFG
1CLEMIAMIANE
2OAKKCGBMIA
3ATLDENBALGB
4GBBALNEBAL
5BUFCLEDENDEN
6SFPHICLECLE
7JACNEKCLAR
8PHIGBLARKC
9KCLARDALDAL
10CHIDALPHIPIT
11TBPITCARPHI
12BALTBINDCAR
13CININDTBIND
14DALSEAPITTB
15NECARTENTEN
16PITNYGBUFBUF
17CARMINNYGNYG
18ARITENMINMIN
19NYJBUFSFOAK
20INDSFOAKSF
21LACARISEASEA
22WASOAKARIATL
23LARATLJACJAC
24HOULACLACLAC
25TENJACNOARI
26DENNOATLNO
27MINDETNYJNYJ
28NYGHOUCHIDET
29DETWASDETCHI
30NOCHICINCIN
31SEANYJHOUWAS
32MIACINWASHOU


Conclusions


[NOTE: As of writing, the odds for the TB/MIN game are still OFF, so take those predictions lightly.]

Let’s ignore Model A for now. It uses only two weeks worth of data and has some crazy ideas about which defenses are good. Since Model C was good for two weeks, let’s keep that rolling along and look at what it predicts for the week.

The top teams this week are driven in large part by the opponent and less by the expected point point totals. The top teams for the week (and their opponents in parenthesis) are: MIA (NYJ), KC (LAC), GB (CIN), NE (HOU), BAL (JAC), and DEN (BUF). Of those, NYJ, LAC, HOU, and JAC are in the top 8 for coughing up points to fantasy defenses over the last 2 years.

It’s likely only MIA and GB will be available to stream out of those top teams, but you have a few more options rounding out the top 12. PHI and LAR are both good choices; good teams, good matchups against LAC and SF, respectively, but the expected point totals for the week aren’t inspiring. PHI may be taken, but LAR is likely available. PIT is hanging out around the bottom of the top 12, and they’re expected to to hold CHI to one of the lowest point totals of the week. You could do worse in a pinch.

I have no idea why DAL is up there this week. They’re not a great defense, and they’ve lost a lot of talent in their secondary this year and ARI doesn’t give up many points to fantasy defenses. I don’t know how they ended up toward the top of some of these lists, but I’d stay away.

What are you doing up there, Cleveland?!? Did I flip some numbers around somewhere… no, wait, you’re favored to win this week against IND who have given up a TON of fantasy points this year. Huh. Let’s mark CLE as a big maybe. Great matchup, and their defense actually scored a lot of points in 2015, but in 2016 (and so far 2017) they have been thoroughly mediocre. This may be the week to get their groove back. Or at least see a little action against a bad Indiana offense.

Week 3 DEF Results

Week 3 K Predictions