2020 NFL Game Predictions: Week 8

   Welcome!  I intend this to be an ongoing project of predicting NFL game outcomes, point spreads, and final scores.  My hope is that my models will improve over time as they become more sophisticated and use better data.  I will try to regularly publish predictions to keep myself accountable, and will make observations along the way about what is working and what isn't.  See below for latest updates.  Enjoy!

Previous 2020 Predictions:

• Week 7: 9-5 (RF); 10-4 (LR)
• Week 6: 8-6 (RF); 8-6 (LR)
• Week 5: 9-5 (RF); 11-3 (LR)
• Week 4: 11-4 (RF); 11-4 (LR)
• Week 3: 7-8-1 (LR)
• Week 2: 8-8 (LR)
• Week 1: Kickoff

________________________________________________________

Recap:

9-5 (LR) and 9-5 (RF).  The LR model had an expected value of 8.879 while the RF model had an expected value of 8.219.  So both models behaved as expected.

Falcons did win despite my personal reservations.  Chargers lose by 1 to the Broncos after giving up 21 points in the 4th quarter.  Dolphins do beat the Rams in accordance with my models and my personal choice.  Ravens lose to Steelers so Steelers remain undefeated.  

Big surprises? Packers lose to Vikings and Titans lose to Bengals.  Ouch.

Average week.  On to the next week.

________________________________________________________

Week 8:

No changes in the models except to retrain them with the latest data.  The models agree on all outcomes this week.  I have noticed that the linear regression model tends to have more extreme probabilities than the random forest model.  I would have thought that it was overly confident except that so far it has been doing a slightly better job predicting than the random forest model.  It will be interesting to see if that continues to be true, or if probabilities that are much higher do tend to win more often (as they should).  Some post season comparison and analysis should yield some interesting results.

14 games this week. Cardinals, Jaguars, Texans, and Washington have a bye.

Oddities and Bold Predictions:

• Falcons vs. Panthers
• Both models choose Falcons but 538 chooses Panthers.  Falcons are 1-6 and Panthers are 3-4.  Falcons previously lost to Panthers and found a way to lose to Lions last week.  I'd personally pick the Panthers.
• Chargers vs. Broncos
• Both models choose Chargers but 538 chooses Broncos.  Both teams are 2-4.  Broncos have played a lot of good teams and beat the Patriots.  Chargers have played relatively good teams as well, kept games close, and are coming off a win.  I think I side with the Chargers on this one.
• Dolphins vs. Rams
• Both models choose Dolphins to beat the Rams.  This is very surprising. 538 chooses Rams, as I would have initially thought.  Dolphins are 3-3 while Rams are 5-2. Both lost to Bills but Dolphins beat 49ers while Rams lost.  Dolphins are coming off a bye week. Surprisingly, after looking at the stats, my gut is leaning towards Dolphins.
• Ravens vs. Steelers
• My models and 538 agree: Ravens victory.  Predicting an end to the undefeated season of the Steelers.

So a difference of 3 predictions compared with 538, and I personally side with my models on 2 of those.  We shall see if that was wise or foolish.  Good luck!

Here are the predictions for Week 8:

2020 NFL Game Predictions: Week 7

  Welcome!  I intend this to be an ongoing project of predicting NFL game outcomes, point spreads, and final scores.  My hope is that my models will improve over time as they become more sophisticated and use better data.  I will try to regularly publish predictions to keep myself accountable, and will make observations along the way about what is working and what isn't.  See below for latest updates.  Enjoy!

Previous 2020 Predictions:

• Week 6: 8-6 (RF); 8-6 (LR)
• Week 5: 9-5 (RF); 11-3 (LR)
• Week 4: 11-4 (RF); 11-4 (LR)
• Week 3: 7-8-1 (LR)
• Week 2: 8-8 (LR)
• Week 1: Kickoff

__________________________________________________________________________

Recap:

10-4 (LR) and 9-5 (RF).  The difference between the two was the Bears-Rams game, which LR correctly predicted a Rams victory while RF incorrectly favored the Bears.  49ers win and Cowboys lose. Seahawks lose to Cardinals in OT, ending their undefeated season so far.  Lions barely beat the Falcons as the Falcons give up the game in the final minutes and seconds.  Steelers beat Titans against my model predictions.  Steelers are the only team still undefeated and Jets are the only team that has not yet won a game.  NFC West is all above .500 while NFC East is all below .500.  

LR had an expected value of 9.723 wins and RF an had expected value of 8.513, so both models performed as expected.  Pretty good week, although disappointed by Seahawks loss.

On to week 8!

__________________________________________________________________________

 

Week 7:

No updates this week.  I may have been kidding myself thinking I could do weekly predictions and make updates and improvements to the models while still doing everything else in my life.  More likely any significant improvements in the data storage, processing, pipeline, modeling, and predictions will have to come in the off season.  But I'll keep trying.

The models agree except in one instance.  14 games this week.  Ravens, Colts, Dolphins, and Vikings have a bye.

Oddities and Bold Predictions:

• 49ers vs. Patriots
• Both models choose 49ers.  But 49ers play away and they have been somewhat inconsistent this season.  Patriots are 2-3 against hard teams and 49ers are 3-3 against average teams.  I'm inclined to go with Patriots personally.  538 chooses Patriots.
• Cowboys vs. Washington
• Both models choose Cowboys but 538 chooses Washington.  It's an away game for Cowboys.  Cowboys are 2-4 and Washington is 1-5.  Cowboys have kept the game closer against better teams.  But they did that with Dak Prescott, who is out for the season.  My source data does not have that data updated, so its not surprising that my models give an overwhelming victory to the Cowboys because they are blind to this fact.  That is a limitation of data driven models: they need good data to be accurate.  Not sure which way I personally go on this one but the models are overestimating the Cowboys at the very least.
• Bears vs. Rams
• LR votes for Rams while RF votes for Bears.  Rams play at home.  Rams are 4-2 and Bears are 5-1.   Bears have had close games against IMHO relatively average teams.  Rams have had close games against better than average teams.  So I don't know, but my gut tells me Rams if I have to choose. 538 chooses Rams.

Good luck!

Here are the predictions for Week 7:

2020 NFL Game Predictions: Week 6

 Welcome!  I intend this to be an ongoing project of predicting NFL game outcomes, point spreads, and final scores.  My hope is that my models will improve over time as they become more sophisticated and use better data.  I will try to regularly publish predictions to keep myself accountable, and will make observations along the way about what is working and what isn't.  See below for latest updates.  Enjoy!

Previous 2020 Predictions:

• Week 5: 9-5 (RF); 11-3 (LR)
• Week 4: 11-4 (RF); 11-4 (LR)
• Week 3: 7-8-1 (LR)
• Week 2: 8-8 (LR)
• Week 1: Kickoff

__________________________________________________________________________

Recap: 

8-6 for both RF and LR.  The expected value for LR was 8.581 and for RF it was 8.06.  So the models performed as expected.

Both of my disagreeing picks with 538 were incorrect, although Lions won as I personally thought they would.  The biggest surprise was the Packers loss to the Buccaneers.  Looks like Packers took an early lead, but with two interceptions that lead to touchdowns, the momentum shifted and the Buccaneers were able to score on offense and shutdown Rogers on defense to a huge victory over the Packers.  The Cardinal's victory over the Cowboys was also pretty crushing, and coincidentally, the final score was the same as the Buccaneers/Packers game: 38 - 10.

On to week 7.

__________________________________________________________________________

 

Week 6:

Somedays just getting the model to run is a challenge.  Due to game schedule changes, changing schemas in data sources, and 'NA' values popping up in new places that caused the models to fail, simply getting the predictions output was no small feat.  So no new changes in the model this week, but at least I have some predictions before game time.  The models agree with each other on every prediction this week, so that's nice.

Only 14 games this week.  Seahawks, Raiders, Saints, and Chargers have a bye.

Oddities and Bold Predictions:

• Jaguars vs. Lions
• My models pick Jaguars but 538 picks Lions.  Lions are 1-3 and Jaguars are 1-4 but playing at home.  Lions have definitely had a harder set of opponents so far but have kept pretty close.  Personally I'd pick the Lions.
• Cowboys vs. Cardinals
• My models pick Cowboys but 538 picks Cardinals.  Cowboys are 2-3 while Cardinals are 3-2.  Cowboys play at home.  Cowboys put up large points but their opponents have put up larger.  I think they have played slightly better teams than Cardinals.  I'm still inclined to go with Cowboys.

All other picks are in agreement with 538, so I guess I didn't mess it up too bad, right? Good luck in week 6!

Here are the predictions for Week 6:

2020 NFL Game Predictions: Week 5

 Welcome!  I intend this to be an ongoing project of predicting NFL game outcomes, point spreads, and final scores.  My hope is that my models will improve over time as they become more sophisticated and use better data.  I will try to regularly publish predictions to keep myself accountable, and will make observations along the way about what is working and what isn't.  See below for latest updates.  Enjoy!

Previous 2020 Predictions:

• Week 4: 11-4
• Week 3: 7-8-1
• Week 2: 8-8
• Week 1: Kickoff

_________________________________________________________________________________

Recap:

9-5 this week using RF model.  Only 14 games played this week as Broncos-Patriots game was pushed back to week 6 and both Packers and Lions had a bye.  Expected value of RF performance was 8.532 correct predictions given the probabilities assigned, so the model performed as expected this week.

Logistic regression model was 11-3, higher than its expected value of 9.016, as the Bears beat the Buccaneers and the Panthers beat the Falcons.  

Browns beat the Colts.  49ers lose to Dolphins in major upset and Chiefs lose to Raiders.  

Overall, happy with the model performance this week as the misses were understandable, and otherwise, the models performed as expected.

_________________________________________________________________________________

 

Week 5:

No changes to the models this week.  What is interesting is that the logistic regression model and the random forest model are giving slightly different results, so it will be interesting to see which one does better.  For the reasons given below, I am inclined to favor the random forest model predictions and will use those to assess this week's results.

Oddities and Bold Predictions:

• Bears vs. Buccaneers
• Logistic model picks Bears to win barely, while RF chooses Buccaneers.  Both are 3-1 and Bears play at home.  One could argue that the Buccaneers have played slightly better teams.  They also have Tom Brady, so I am personally inclined to go with them.  538 agrees with a Bucs win.
• Colts vs. Browns
• Both models pick the Colts.  Browns play at home, and after an exciting and energizing game last week, they are going to be pumped up for this next game.  Both are 3-1.  Both have played about the same level of teams in my estimation.  I'm kind of leaning towards the Browns, but maybe that is because I want them to be good after so long of being bad.  538 has a 50-50 split that picks the Browns.
• Falcons vs. Panthers
• The RF picks the Falcons while the logistic regression model picks the Panthers.  Falcons play at home.  Panthers are 2-2 and Falcons are 0-4.  Falcons have had a hard schedule so far, harder than the Panthers in my opinion, but have put up good numbers.  I could go either way personally.  538 picks the Falcons.

Good luck in week 5!

Here are the predictions for Week 5: