Week 6 of the college football season is in the books and now we can get a look at how well Forward Progress predicts game outcomes in the wild.

This week saw a few upsets, some bigger than others. Let’s dive into this week’s most surprising games and see how well Forward Progress stacked up against the AP Poll.

  • (17) Louisville - 25 vs. (24) NC State - 39
    • This wasn’t the biggest upset of the weekend but NC State outperformed the AP Poll’s expectation by dropping so many points on a team 7 spots ahead of them in the poll. Forward Progress had these two teams nearly even, with NC State expected to score 29.4 points against Louisville’s 29.1. A tiny win against the AP poll, but I’ll take it!
  • (NR) Iowa State - 38 vs. (3) Oklahoma - 31
    • This was definitely the biggest shock on Saturday as the Cyclones beat the Sooners in Norman. Iowa State hadn’t beaten OU since 1990, longer than any of the players on either team have been alive. My prediction model agreed with everyone else and favored OU by 14 points but upsets like these are what keeps the sport exciting.
  • (NR) Michigan State - 14 vs. (7) Michigan - 10
    • Nasty weather made this game a true outlier as both teams were forced to adopt new gameplans in the face of a torrential downpour. Michigan State passed for only 94 yards but ended up with 3 interceptions against the Wolverines, leading to a victory for the unranked Spartans. Forward Progress agreed with the AP poll in this matchup, expecting Michigan to score 29.8 against MSU’s 18.4. I’m absolutely willing to chalk this error up to the weather.
  • (NR) Stanford - 23 vs. (20) Utah - 20
    • I didn’t see any of this game but the highlights show a close competition, with Stanford going up early and Utah unable to mount the comeback. The prediction model agrees with reality in this case, showing Stanford scoring 31.7 to Utah’s 28.7
  • (NR) LSU - 17 vs. (21) Florida - 16
    • Finally, LSU bounced back from a disasterous upset last week against Troy to beat Florida. Forward Progress predicted a close game, 23.4 LSU vs. 21.4 Florida but was correct in the end.

When it was all said and done, my model called the upsets more often than not, going 3/5 against the AP poll. Other notable misses were a predicted upset of Florida State over Miami and Minnesota vs. Purdue. In total, the model held steady with previous training accuracy at 72.55% and 0.447 Kappa, correctly predicting 37 out of 51 games.

On to this weeks rankings!


  Team Predicted Wins Ranking Date Rank Movement
1 Alabama 245 2017-10-08 0
2 Clemson 243 2017-10-08 0
3 Ohio State 239 2017-10-08 0
4 Penn State 239 2017-10-08 1
5 Oklahoma 238 2017-10-08 -1
6 USC 236 2017-10-08 0
7 Miami (FL) 230 2017-10-08 6
8 Washington 230 2017-10-08 -1
9 Wisconsin 229 2017-10-08 2
10 Florida State 228 2017-10-08 -1
11 Notre Dame 225 2017-10-08 3
12 Virginia Tech 223 2017-10-08 -2
13 Georgia 221 2017-10-08 3
14 Auburn 219 2017-10-08 -6
15 Michigan 218 2017-10-08 -3
16 Oklahoma State 217 2017-10-08 -1
17 Michigan State 212 2017-10-08 11
18 Washington State 211 2017-10-08 0
19 Texas Christian 209 2017-10-08 1
20 North Carolina State 205 2017-10-08 6
21 Colorado State 203 2017-10-08 0
22 Georgia Tech 203 2017-10-08 -3
23 Stanford 203 2017-10-08 0
24 West Virginia 198 2017-10-08 -7
25 Iowa 195 2017-10-08 7

The top 3 remain unchanged from last week. Oklahoma changes places with Penn State and falls out of playoff contention on the back of their loss. Depending on how the Sooners play during their next games I doubt they will slide much further but the Iowa State loss certainly hurts. Several teams experienced big movement on the back of their week 6 performances, including Miami, Michigan State, NC State, West Virginia, and Iowa. Auburn also fell a number of slots after a fairly convincing win over Ole Miss. Given the model’s 10 game lookback a single game should not have that much impact on the standings and the 6 spot drop likely has more to do with the tight competiton in the 5-15 groups.

I’m happy to see that the model is performing as expected and furthermore making meaningful changes to the standings based on weekly results. As the season progresses I’ll be working on charting rankings over time and hopefully work up good visualizations for digging into what is driving the rankings in the model.