march-machine-learning-mania-2026

[manic echo]

We're using the same evaluation format as last year. I personally am OK with that, but what does everyone else think?

[thick vessel]

this is my first time to joined any coding Competition

[spring prawn]

Same! Does anybody have any tips on how to get started?

[void harbor]

Logistic Regression: Surprisingly effective for sports because it handles win probabilities naturally.

[lone pecan]

anyone know where we could get player-level data for free, and/or team NIL data?

[eager turtle]

Hi! I was working on this problem statement today and achieved some decent result using Feature Based Bradley Terry Implmentation using Logistic Regression
(notebook link for reference: https://www.kaggle.com/code/rupankarmajumdar/march-machine-learning-mania-2026-bradley-terry)
Would love to hear your ideas on it, and how it can be improved. Next ig I'm gonna focus on some feature builders like elo, massey, seeds, efficiency etc with ensemble learning using Light GBM.
(Feel free to use this as a baseline or for reference)

[void harbor]

wait - we have to find the data? im confused

[lone pecan]

no i just thought id be interesting to see how much that impacted my model.

currently running an ensemble approach using XGBoost, Logistic, Enet, mlp and a simple elo logistic. though it seems past year winners didnt use nearly that much complexity....

[void harbor]

I see - sorry my bad. I know nothing about usa college football I live in UK

[turbid robin]

You're not going to be able to find consistent free NIL data. The schools themselves keep that under wraps because they can. As for the model itself -- this competition is really easy to overfit with complicated models and features. I'd probably use the simplest model I can even if it's "just" XGBoost or Logistic Regression.

[novel quest]

Newbie to basketball here 😄

Does anyone know which datasets will be updated before the final submission deadline?

Will MRegularSeasonDetailedResults be updated?
And which datasets will not be updated?

[swift prairie]

Complete newbie here
If you don't mind could you pls explain what you mean by updated? Why would they update the datasets?

[swift prairie]

"File descriptions
Below we describe the format and fields of the competition data files. All of the files are complete through February 4th of the current season. As we get closer to the tournament in mid-March, we will provide updates to these files to incorporate data from the remaining weeks of the current season."

nvm

[limpid flume]

and nevermind, I see the post in the discussion forum

[spark prawn]

Hey everyone! Quick question: is the submission just the prediction data, or do we need to upload the model code too? Thanks!

[manic echo]

Just the prediction data, but if you score very well, they may ask for your notebook

[spark prawn]

Thanks, Thomas! Appreciate the quick answer. Hopefully, I'll actually be in a position where they want to see my notebook! Thanks again!

[lusty mauve]

has the dataset been updated yet? How will I know if it has been updated? I've been checking the competitions page on kaggle regularly

[turbid robin]

Dataset won't be updated until Monday most likely. The official bracket isn't out yet for a few more hours.

[lusty mauve]

Thanks

[turbid robin]

Brackets are ready now if you haven't checked yet

[mortal meteor]

I just connected my kaggle so I can post. Wish I did it sooner I almost missed the deadline.

[glass kernel]

Post?

[manic echo]

What's wrong with the leaderboard?

[lusty mauve]

Are those scores really that low?

[manic echo]

They're not being evaluated correctly

[barren pollen]

I was being evaluated now my submission marks error. Does anyone have the same problem? Just moments ago my submission was marked ok.

[white timber]

same issue here

[glass kernel]

The lower the score the better

[glass kernel]

Probably the submissions were being invalidated and rescored again using the corrected data

[glass kernel]

0 is best while 1 is worsg

[lusty mauve]

But 0?? A perfect 0 is possible? I didn't even know that 😅

[glass kernel]

I mean if u get all the predictions right haha

Which is really unlikely tho

[lusty mauve]

True 🫡

[hard nimbus]

https://www.kaggle.com/competitions/march-machine-learning-mania-2026/leaderboard?search=ddan

#7 Rank with winning + Gold medal out of 3114 competitors on Kaggle

[glass kernel]

There's still more games tho

[hard nimbus]

yes correct

[hard nimbus]

how many left? do you know

[glass kernel]

For round 1 I think it's 110

[turbid robin]

Half of the first round of the men played yesterday (16 games). 110 left. PxlPau’s math is correct

[glass kernel]

Woohoo!

[glass kernel]

Wait edit it was for all the remaining rounds including finals

[turbid robin]

Yeah 110 games total left for both tournaments in all rounds

[sacred tide]

How many games are played and how many are left?

[livid mirage]

Hello! There is exactly 104 games remaining both men and women combined.

[sacred tide]

How many were played?

[livid mirage]

30 were played accross both men and women.

[sacred tide]

Damn this leaderboard fluctuates a lot 😅

[livid mirage]

That is true, a better logical metric is to observe the leaderboard as it approaches the finals.

[glass kernel]

Oh no Texas defeated Gonzaga my prediction is cooked 😭

[hard nimbus]

hehe now few are left 16 to 8 and then 4 in those final dates 5 to 7 april

[glass kernel]

😭 too many upsets

[sacred tide]

Yes

[pale dome]

why is submission closed for this?

[glass kernel]

The matches are being played out

[soft sun]

https://www.linkedin.com/posts/pierre-raffalli_machinelearning-kaggle-ncaa-ugcPost-7447291574503366657-8KSX?utm_source=share&utm_medium=member_ios&rcm=ACoAABKrYTMBvlwIBSa4_2vuhwc-Sf9aGASGY4c

[warm glacier]

check my bio 😁

[clever heart]

check my bio 😁