Intro
In my opinion, March Madness is probably the best sporting event of the calendar year. This first weekend is loaded with fun games and upsets, and every year I like trying to see if my college basketball model can win a bracket pool (spoiler: it can't).
My first attempt at this was during my junior year in college as a CS student. I created a computer generated bracket that almost won the Red Shed pool of a couple hundred people, but lost because Texas Tech lost in the finals. Bummer.
This algorithm was super straightforward, I think just randomly picking winners weighted by seed. Every year since then, I have tried to make the models more sophisticated. I was away for the past 5 days on a golf trip (Bandon review blog coming soon) so I am last minute scrambling to fill out my brackets this year. Since I am running super late on this post, giving a few takeaways below from the 10,000,000 sims I have run.
The Model's Takeaways
- Model loves Houston
Of the 10 million sims run, the model picked Houston to win 7.8 million of them. This is an absolutely insane number, I have no idea why this is the case. I think that it values defense really highly, but if that is the case then I don't know why it respects Alabama so much. Either way, might be some value on Houston since I feel like they aren't as talked about nationally as some of the other top seeds.
- McNeese to take down Vandy
Pretty much every year a 5 seed loses to a 12, and the model is calling its shot that it will be Vandy. Current spread of this game is Vandy -11.5, so it is a bold take. We'll see how it plays out for him.
- Tennessee to make a run
Similar to how it values Houston's defense, the model thinks quite highly of Tennessee. It gives them the third (!) most likely odds to win it all. Can't say I agree with this one but I appreciate the hot take.
- Virginia shaky as a 3 seed
The model gives Wright State a relatively decent chance at upsetting Virginia. This is an 18.5 point spread, would be a hilarious upset.
- 10 seeds to start strong
The model has both Mizzou favored over Miami and Texas A&M favored over Saint Mary's. Not crazy upsets by any means but it feels relatively confident on these.
- UCLA over UConn, BYU over Gonzaga
If these teams meet in the second round, the model is picking the higher seed more often than not.
Simulations
The 10 million simulations resulted in the below winners. I always think it's funny to run these then look at the bracket where a team like Akron wins. I guess they have a 1 in 10 million shot.
| Team | Championship Wins |
|---|---|
| Houston | 7,881,051 |
| Florida | 993,769 |
| Tennessee | 579,905 |
| Duke | 345,934 |
| Alabama | 93,143 |
| Arizona | 39,437 |
| Purdue | 32,882 |
| Michigan | 6,792 |
| Kentucky | 6,640 |
| Wisconsin | 5,871 |
| Texas Tech | 5,744 |
| Michigan State | 4,516 |
| Brigham Young | 2,784 |
| Gonzaga | 354 |
| St Johns Ny | 251 |
| Kansas | 242 |
| Iowa State | 175 |
| Arkansas | 102 |
| Louisville | 94 |
| North Carolina | 70 |
| Central Florida | 68 |
| Missouri | 46 |
| Ucla | 38 |
| Texas Am | 23 |
| Utah State | 14 |
| Villanova | 10 |
| Connecticut | 8 |
| Clemson | 8 |
| Saint Marys Ca | 5 |
| Nebraska | 5 |
| Southern Methodist | 5 |
| Texas | 4 |
| Virginia Commonwealth | 4 |
| Mcneese State | 3 |
| North Carolina State | 1 |
| Georgia | 1 |
| Akron | 1 |
The Bracket
Here is the model's official bracket for 2026. I honestly don't hate its final four, though I think it grossly overrates Tennessee. I'm fully ready for this post to age terribly if Houston gets bounced early but that is kind of the fun of it. Best of luck everyone, this tournament is a blast.
