Data scientist, physicist, and fantasy football champion

Week 16 DEF Predictions

Welcome to the final week of the Fantasy Football playoffs!

Hopefully many of you reading this are in your respective championship games. I’m in a championship game… of sorts. The Toilet Bowl: the game for last place. To be fair though, it has not been because of the performance of my defense or kicker. The rest of my team sucks and is all the more reason to spend the offseason figuring out how to predict QBs and TEs (RBs and WRs are a dream at this point, but who knows).

Last week every model did okay. I would say that we’re at a point where models can accurately predict things, but Week 14 was a hot mess, so it’s probably a bit of a statistical fluke that they did as well as they did.

Two administrative notes: First, I’m working on an R package now to make my life easier and more quickly create new models. If anyone has any interest in seeing it please let me know. Second, I will most likely not be predicting defenses and kickers for Week 17. If there is a request I will happily run the model, but otherwise this will be the last prediction post of the season. For those of you who want to write to me, my Twitter account is @FFDataStream. Tweet at me if you want any of these things.

On to the models:

Model A:

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 2016
  • Best model from weeks 8 and 9, 3rd best in week 10
  • Terrible in week 12 but great in Week 13

Model B:

Model B parameters:

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

Model B Notes:

  • Uses only data from 2016
  • Comparable results to Model A while reducing overfitting
  • Terrible in week 12

Model C:

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 and 2016
  • OK performance weeks 8 and 9, best model in week 10

Model D:

Model D 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 D Notes:

  • Uses only data from the last 6 weeks

Model E:

Model E 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
  • NEW: “Team:week” term (does the team have momentum?)
  • NEW: “Opponent:week” term (does the opponent have momentum?)
  • NEW: “week” term (basically a bookkeeping term if I want to use the other two)

Model E Notes:

  • Uses data for last 8 weeks
  • includes week + Team:week and Opponent:week terms for each team
  • New terms try to account for momentum

Model F:

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 and 2016
  • Best model in weeks 11*, 12, and 13
  • Ignores defense TDs but weights interceptions and fumbles to try to account for it

Week 16 model summary

Rank

Model.A

Model.B

Model.C

Model.D

Model.E

Model.F

1

KC

KC

KC

KC

KC

KC

2

SD

SD

NE

SD

SF

NE

3

NE

NE

SEA

TB

BAL

SD

4

SEA

BUF

SD

NYG

DAL

SEA

5

SF

SEA

TEN

SF

SD

NYG

6

NYG

ATL

NYG

TEN

ARI

TEN

7

BUF

SF

DEN

GB

NE

GB

8

ATL

CLE

ATL

NE

ATL

LA

9

GB

MIN

LA

ATL

GB

DEN

10

CLE

GB

GB

DAL

PHI

PIT

11

MIN

TB

HOU

OAK

OAK

HOU

12

TB

PHI

PIT

SEA

CHI

OAK

13

PHI

NYG

WAS

ARI

TEN

BUF

14

DEN

LA

PHI

CLE

WAS

ATL

15

PIT

MIA

ARI

PIT

CAR

PHI

16

TEN

PIT

SF

MIA

NYG

ARI

17

MIA

TEN

MIN

DEN

TB

SF

18

LA

OAK

BUF

LA

PIT

CLE

19

ARI

DEN

CLE

PHI

CLE

MIA

20

OAK

CAR

OAK

HOU

MIA

CIN

21

WAS

WAS

MIA

CAR

IND

WAS

22

HOU

ARI

JAC

WAS

NO

DAL

23

CIN

HOU

TB

MIN

HOU

CAR

24

CAR

CHI

CIN

CIN

DEN

BAL

25

CHI

JAC

CAR

BAL

NYJ

TB

26

JAC

CIN

DAL

CHI

DET

NO

27

BAL

NO

DET

NO

MIN

MIN

28

NO

DAL

NO

DET

CIN

JAC

29

DAL

BAL

BAL

IND

JAC

DET

30

DET

DET

CHI

BUF

BUF

CHI

31

IND

IND

IND

JAC

SEA

IND

32

NYJ

NYJ

NYJ

NYJ

LA

NYJ

KC, NE, SD, and SEA top pretty much all of the lists this week. Of those, likely only SD is available, and even then there’s not a great chance of that. Play (or pick up and play) any of these with confidence.

Model B matches Model A, Model C matches Model F, and Models D and E are mostly garbage, so le’ts compare Models A and F. After that initial group there are not a lot of similarities between Models A and F. I haven’t been paying any attention to the Giants this year, but they’ve been doing quite well and have scored the 6th most points overall (by my league). They’re up against the Eagles this week (who are in the middle of the pack for points given up to defenses) in what is predicted to be a low-scoring game. By comparison, Tampa Bay has a similar number of fantasy points on the season against NO who have a similar record against defenses as Philly. That game is expected to be a shootout, though, with the total number of points expected for that game 12 points higher than the Giants/Eagles game. I’d skip both NO and TB this week if I were you.

I like TEN against JAC for al almost-certain 4 points, but I would doubt much more than that. I would say the same for GB against MIN. MIN has been very stingy with the turnovers but has scored 20 or fewer points in 10 of 15 games and tends to give up 2 or 3 sacks a game.

There’s not much else that A and F agree on. The LA/SF game is actually a great example of the difference between the models. The game is expected to be very low scoring (40 points total) and LA has given up a lot of points to defenses, but SF has not picked up many points all season. I’d definitely stay away from SF but I’ll consider LA. I just really like Model F I guess.

This week should be a good final exam for my models. Again, I won’t be back next week with a prediction unless there’s an explicit request, but I will start writing more data analysis articles and in the next few weeks hopefully I’ll get a chance to write up a season summary of all of my models.

Good luck future champions!

(That message was for everyone but Ben)

Week 15 K Evaluation

Week 15 DEF Evaluation