Aike’s Model vs. Kenpom

[Aike]

Haven’t done this in awhile. Decided to go back and track performance of my model vs. Kenpom in predicting NCAA Tournament game winners.

Looked at 2022 and 2023 so far. Here is the fairly simple analysis:

If you knew before the tournament who would play in each of the 63 games, who would you pick to win each game before the start of the tournament?

This is a little more fair, because neither model gets dinged as much for early upsets.

For example, when Purdue lost to FDU in 2023, both models missed that one. But they both would have correctly predicted that FAU would beat FDU in the second round.

Here were the results:

2022
KP - 38 correct
Aike - 41 correct

2023
KP - 43 correct
Aike - 44 correct

 

[Aike]

Note that this does not consider any other team outside of the two teams playing each game. Doesn’t factor in that one team may start the tournament with a better draw, etc. Just which team was rated higher before the tournament started.

 

[Dead Cat Bounce]

Suck it, Kenpom.

 

[St.PatterSoN-54-]

Faaaaaaaaar Ken and his poms

 

[Aike]

I’ve always said, and I think this sort of backs up that a good strategy for filling out your bracket would probably be to take my model, Kenpom, and one other that you like.

Pick the team that at least 2 of 3 models agree on, but maybe do a little digging on any extreme outliers.

 

[Aike]

I checked 2024 - Kenpom won that round 45 to 44.

KP was 15/16 in round 2 last year, vs. 13/16 for me.

The overall results flipped on UT/Purdue. I had UT rated slightly ahead of Purdue. That “win” for KP flipped the results in his favor.

Across the past 3 years, I was 129/189 vs 126/189 for KP. I averaged beating him by 1 game per year. 68.2% vs. 66.7% success rate.

For the record, I didn’t personally pick UT to make the FF, despite the model predicting it. Just didn’t trust them to stop being Tennessee.