March Madness 2025 Model

We spent the last few days building a new type of model for the March Madness Tournament this year. Our results were submitted to the Kaggle March Madness Machine Learning Mania Competition for the 2025 Tournament. The winner gets $10,000 and testing our model on the past few years shows competitive performances every year, with a theoretical second place finish in 2023.

The model is based on the Elo rating system with two key changes. First, each team doesn’t start the season with the same Elo rating. Rather, we use a very specially chosen prior to give the model a head start. This allows us to use a smaller “K-value” and have more stable ratings. Second, our model takes into account margin of victory in a new way. That means that a 20 point win nets more points than a 2 point win. A heavy favorite that wins by only a couple points may actually lose rating as a result, a stark contrast from existing models that incorporate margin of victory into Elo ratings. I will write more about these techniques in detail if the model does well in the competition. If not, the point is moot anyway.

Here are the ratings we came up with for the march madness 2025 tournament. It turns out that our model thinks, far and away, that Florida is the team to beat and stands the best chance of winning the whole thing. Not only that, but it thinks that Duke is closer to #5 Tennessee than they are to Florida.

Instead of deep diving every matchup like we sometimes do, I want to point out some interesting observations.

• The talk of the media was that Louisville was a huge snub, being 10th in the rankings and an 8 seed. However, our rating system thinks Gonzaga is the bigger snub! They edge Louisville by a couple points.
• Our model thinks that Purdue is one of the most overrated teams in the tournament, barely top 25 but snagging a fourth seed. They could get upset early. But they aren’t nearly as big of an overseed as Memphis.
• Lots of discussion about North Carolina not deserving to be in, but they actual rank as the second best of the 11s, just behind VCU.
• Missouri might be way underrated
• Best team at each seed:
  • 1: Florida
  • 2: Tennessee
  • 3: Wisconsin
  • 4: Maryland
  • 5: Clemson
  • 6: Missouri
  • 7: St. Mary’s
  • 8: Gonzaga
  • 9: Georgia
  • 10: Arkansas
  • 11: VCU
  • 12: Liberty
  • 13: High Point
  • 14: Lipscomb
  • 15: Robert Morris
  • 16: Norfolk St.
• Of the top seeds, the worst at each seed:
  • 1: Houston
  • 2: St. Johns
  • 3: Kentucky
  • 4: Purdue
  • 5: Memphis
  • 6: Mississippi St.
• Quick picks:
  • Cinderella teams: (double digit seed) Arkansas (single digit seed) Gonzaga
  • Team outside top 4 that could win: Missouri
  • Most likely early upset: Kentucky and Purdue

Have fun with the tournament this year, and I hope these rankings help!

To receive email updates when new articles are posted, use the subscription form below!