Data scientist, physicist, and fantasy football champion

Week 7 DEF Predictions

Last week many of the defenses did better or worse than predicted, but yet again it was Model F that among the top models along with Yahoo’s experts. This week I’ll use Model F again to pick a defense. I still haven’t modeled the effect of a replacement QB yet, so those of you looking to stream against GB or TB (maybe) may be looking at a gold mine or maybe not. I’ll work on this, because I can’t just go through life assuming that prevailing wisdom is correct. I need data, dammit.

Let’s see some predictions for this week:

ModelTypeScoreOppScoreOppHomeAwayDataYearsPointSystem
ALMXXXX2017IOUH/Standard
CLMXXXX2015-2017IOUH/Standard
FLMXXXX2015-2017TruScor
GGibbs SamplerXXXX2015-2017TruScor
HLMXXXX2017TruScor
IMany LMsXXXX2015-2017IOUH/Standard

 

Model A:

.Model A-1.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


Model H:

.Model H-1.png


Model H 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 H Notes:

Uses only data from 2017
Uses TruScor points to reduce the effect of TDs


Model I:

.Model I-1.png


Model I parameters may include:

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 I Notes:

Data from 2015-2017
This model tries to predict all of the components of scoring separately (interceptions, sacks, etc.). Each component comes from a linear model containing only the significant terms (p < 0.05) from the list of possible terms above.

Week 7 model summary:
 

RankACFGHI
1JACMINLACTENNOMIN
2NOARITENCARPITMIA
3PHICARCARLACJACCAR
4PITJACMINSEALACJAC
5WASLACPITPITPHILAC
6MIAMIAMIAMINMIASEA
7SEADENJACMIAMINDEN
8CARTENARIJACTENTEN
9TBPHISEANOCARPHI
10LACSEANOARISEAARI
11BALNODENDENWASPIT
12NYJLARLARPHINYJNO
13INDKCPHILARTBKC
14BUFPITTBKCBALLAR
15CHITBCHITBARICHI
16MINCHIKCBUFCHIBUF
17KCNYGBUFDALINDTB
18LARINDNYJCHIBUFIND
19CINCLENYGNYJDALNYG
20TENBUFINDNYGDENCLE
21ARIBALDALINDKCDAL
22DENWASWASBALLARNYJ
23NYGDALBALCINNYGCIN
24DALNYJCINWASNEBAL
25GBATLCLENECINWAS
26SFCINOAKOAKSFNE
27NEOAKNECLEGBATL
28CLENESFSFCLEOAK
29OAKSFGBGBOAKSF
30ATLGBATLATLATLGB


Conclusions


This week Model F shows a big group of 11 defenses at the top, 3 in a mid-tier, then 10 more all grouped together. This roughly matches the predicted point totals for the week; no 12 point favorites and no super-low-scoring affairs. The Giants have th elowest predicted score this week (16.25 points) against Seattle. Seattle would be a good play (in that top group), but they’re the kind of defense that someone has hoarded, thinking they’re worth playing every week. MIN, SEA, JAC, DEN, and KC will be taken, but for the rest of you here are some teams in that top group that you might be able to grab:

  • LAC (against DEN): Predicted scores aren’t especially low, but DEN has scored 16 or fewer points in their last 3 games and give up 3 or 4 sacks, an interception, and a fumble per game. With 2 of their top receivers injured (out or at least questionable) this is probably my top choice for the week. The Chargers have been allowing a lot of points but have been racking up sacks.
  • TEN (at CLE): Always play a defense against Cleveland. It might be Kizer, it might be Hogan, but either way expect a lot of turnovers. TEN hasn’t done well this year, but they won’t have to against Cleveland
  • CAR (at CHI): CAR has a lot of sacks but not a lot of fantasy points this year. CHI fumbles often, though, and the rookie QB (Trubisky) is bound to make a few mistakes. Not a bad choice.
  • LAR (vs. ARI): ARI gives up sacks and turnovers. The Rams have scored the 4th most fantasy points this year. Another great pick for the week

And some cautions: 

  • ARI (at LAR): I really don’t know why this one is up here. Goff hasn’t turned over the ball that much, and the offensive line is doing a good job. This one might be around for historical reasons: for the past 2 years ARI has been a great defense, but this year they haven’t done much. Of the teams in the top tier, I’d avoid this one. 
  • BUF (vs TB): Buffalo has been great this year. No opponent has scored more than 20 points, they get about 2 interceptions per game, and they’re toward the top of fantasy points scored despite no TDs. Still, TB doesn’t turn over that much. Check how Jameis Winston is doing in practice this week. If he’s out this is a FANTASTIC play, but if he’s in I’d avoid it.

Good luck to everyone. And if you’re the kind of person who is, say, counting on Denver to help them out this week, just know that my team is so much better that you have no chance. Hypothetically.

Week 7 QB Predictions

Week 7 K Predictions