Machine Learning College Predictions.
Each game predicted.
Statalysts fits a single power rating to every college football and basketball team from the season's scoring margins, anchored to a preseason projection and pulled toward each team's own results as the evidence accumulates. Those ratings produce a projected margin and win probability for every upcoming game. Want to create your own matchup? Open the custom predictor and see how the model rates any two teams.
Kansas State
77.1%
Colorado
22.9%
The model makes Kansas State a 8.5-point favorite, which is 10.0 points clear of the market's line. This is the largest gap between the model and the market on this week's board.
Week 9 recommended games
The games where the model is at least five points away from the posted market line. A wide gap is a measure of disagreement. Click any game for the full breakdown.
| Matchup | Model line | Market line | Edge | Win prob | |
|---|---|---|---|---|---|
|
Kansas State
4-3
Colorado
2-5
Sat, Oct 31 · 2:00 PM ET
|
Kansas State −8.5 | Colorado -1.5 | 10.0 | 77.1% | |
Tulane
3-4
Charlotte
1-6
Fri, Oct 30 · 7:00 PM ET
|
Tulane −18.1 | Tulane −9.5 | 8.6 | 93.0% | |
2
Ohio State
6-1
USC
6-2
Sat, Oct 31 · 7:30 PM ET
|
Ohio State −14.5 | Ohio State −6.5 | 8.0 | 88.9% | |
Marshall
4-3
Old Dominion
2-6
Sat, Oct 31 · 1:00 PM ET
|
Old Dominion −4.5 | Marshall −3.0 | 7.5 | 65.4% | |
UTEP
1-6
North Dakota State
6-2
Sat, Oct 31 · 3:30 PM ET
|
North Dakota State −32.0 | North Dakota State -24.5 | 7.5 | 99.0% | |
UAB
4-3
South Florida
5-2
Sat, Oct 31 · 12:00 PM ET
|
South Florida −14.8 | South Florida -7.5 | 7.4 | 89.3% |
No team by that name
Nothing on this slate matches “”. Check the spelling, or clear the search.
How the recommended games finished
Every recommended game, a 5+ point edge over the market line, from the most recently completed week, graded against the spread and on the outright winner. Click any game for the full breakdown.
Last week
2026 season
Pushes are listed last in the spread record and don't count toward the percentage.
| Date | Final score | Model called | Recommended | Outcome | |
|---|---|---|---|---|---|
| 2026-10-24 |
Indiana
13
Michigan
6
|
Indiana by 15.190% win prob | Indiana 6.6 pt edge |
Winner
No cover
|
|
| 2026-10-24 |
Iowa
10
Minnesota
13
|
Iowa by 12.786% win prob | Iowa 7.2 pt edge |
Winner
No cover
|
|
| 2026-10-24 |
North Texas
49
Navy
39
|
North Texas by 8.677% win prob | North Texas 7.1 pt edge |
Winner
Covered
|
|
| 2026-10-24 |
NC State
28
Stanford
31
|
NC State by 13.387% win prob | NC State 5.8 pt edge |
Winner
No cover
|
|
| 2026-10-24 |
Army
38
Tulsa
35
|
Army by 4.465% win prob | Army 5.9 pt edge |
Winner
Covered
|
|
| 2026-10-22 |
James Madison
31
App State
34
|
James Madison by 18.794% win prob | James Madison 8.2 pt edge |
Winner
No cover
|
Top ten, both sports
A rating is a team's strength in points. The difference between two ratings is the model's projected margin on a neutral floor.
Ohio State
Notre Dame
Miami
Indiana
Alabama
Georgia
LSU
Texas
Utah
Oregon
Michigan
Gonzaga
Arizona
Iowa State
Purdue
Duke
Vanderbilt
Illinois
BYU
Louisville
How a rating becomes a prediction
Collect the season
Every completed game is pulled from the college football and basketball data APIs: final scores, schedule, venue and, where posted, the market line. Football rates FBS and FCS teams on the same scale, because a large share of FBS games are played against FCS opponents and those results only mean something if both sides carry a real rating.
Fit a rating to each team
A ridge regression fits one rating per team to the season's scoring margins, so a team's number reflects who it played as well as by how much it won. Early in the season there is little to fit on, so the regression is pulled toward a preseason projection built from the team's own final rating last season and a published preseason rating.
As games accumulate, the weight on this season's results grows and the prior fades.
Project the game
The difference between two ratings, plus home advantage where it applies, is the projected margin. That margin converts to a win probability for each side.
Where a sportsbook has posted a line, the model's number is compared against it. The gap between the two is the edge — not a claim that the model is right, but a measure of where it disagrees most with the market.