Data scientist, physicist, and fantasy football champion

Week 12 DEF Evaluation

This week I’m going to evaluate six models. That’s right, six. Did you all find the surprise model I hid in the “Medthodology” section of this site? No? You mean you don’t check every page for updates, especially on holidays?

The first five models are all the same as last week and can be found easily. Models A and B use 2016 data only, Model C uses 2015 and 2015, and Models D and E use only part of 2016 with a focus on only the last few weeks (to account for hot streaks).

The new one, Model F, is a weird one. It uses the same factors and data set (2015 and 2016) as Model C, but it ignores TDs and adds extra weight to fumbles and interceptions. The idea being that predicting TDs is tougher than predicting picks and fumbles, so I tried to guess how often to expect a TD from a turnover and weight them accordingly. I call this system TruScor, and whenever someone explains to me what a trademark is I’m totally trademarking it. For now, though, we’ll have to go with the less formal ‘dibs’. Check out the post from Thanksgiving to read more about the methodology (hint: you’ll wish you did when you seee the results).

Let’s see how these models did in week 12:

Rank

Model.A

Model.B

Model.C

Model.D

Model.E

Model.F

FantasyPros

DWC

True.Order

True.Score

1

ATL

PHI

KC

PHI

ATL

NYG

NYG

BUF

NYG

23

2

PHI

BUF

NYG

NYG

PIT

KC

BUF

NYG

TB

19

3

BUF

ATL

DEN

BUF

MIA

DEN

DEN

BAL

KC

15

4

HOU

HOU

BUF

ATL

SD

BUF

MIA

SEA

SD

13

5

KC

SD

PHI

PIT

KC

OAK

TEN

DEN

OAK

11

6

NE

NE

PIT

HOU

TB

PHI

KC

TEN

PIT

11

7

SD

KC

ATL

BAL

BAL

PIT

SEA

KC

GB

10

8

DEN

MIA

NE

SD

DEN

ATL

NE

NE

BAL

8

9

CHI

PIT

SEA

NO

TEN

NE

BAL

MIA

NO

7

10

NO

NO

HOU

KC

BUF

HOU

NO

NO

CIN

7

11

PIT

NYG

OAK

SEA

NYJ

SEA

SD

DET

SEA

7

12

NYG

CHI

BAL

MIA

CLE

BAL

PHI

SD

MIA

6

13

BAL

DEN

NO

DET

CAR

NO

MIN

ATL

NE

6

14

MIN

MIN

SD

TB

DET

SD

PIT

PHI

DET

6

15

MIA

BAL

CHI

NE

DAL

MIA

HOU

MIN

DEN

5

Model.A

Model.B

Model.C

Model.D

Model.E

Model.F

FP.top20

DWC

98

96

64

79

67

50

65

70

Unsurprisingly, D and E are the only models that actually placed TB in top 15 this week since they weight recent performance more heavily than the other models. Quite surprisingly, TB did great last week. They were one of the top defenses this week (as they were two weeks ago), this time taking 2nd place with 19 points and making Models D and E seem a little better than I think they are. If you ignore TB, Models D and E both score around 80, much worse than FantasyPros or Model C. I keep waiting for solid proof that shows I can just drop these two models and clean up these tables a little, but that proof doesn’t exist yet. Some day soon, maybe. Or maybe TB really is good and I’ve been underestimating them. Either way, expect 6 models for Week 13.

For the last 2 weeks I’ve been liking Model C more and more, and sure enough, it once again beats both the average of the top 20 FantasyPros experts and Defense Wins Championships. However, it doesn’t quite blow them out of the water. A slight improvement in accuracy over other people’s models is not nothing, but had one or two plays gone differently and DEF scores shifted just a little bit I could easily see Model C beaten by FantasyPros.

The real winner this week is the newcomer, Model F. Take a bow, Model F! This model is an improvement on Model C (which I have already been trusting more and more) and was the best this week and would have been the best last week had I used it. I plan to trust Model F this week, and I recommend you do the same.

Or, if you’re Carly, don’t bother because it won’t help you anyway.

Week 13 DEF Prediction

Week 12 K Prediction