/Kaggle Submission · March Machine Learning Mania 2026
Our
Predictions
We predicted the winner of every possible game in both the men's and women's NCAA tournaments — 132,133 matchups total — and submitted them to a real data science competition on Kaggle.
💡 What is this page?
Imagine asking a computer: "If Duke played every other team in the tournament, how often would Duke win?" We did that for every single pair of teams — not just the games that will actually happen, but every possible matchup. Instead of guessing, our computer uses real stats from this season to calculate a win percentage for each game. A prediction of 75% means "if these teams played 100 times, we think Team A wins about 75 of them."
/Model Architecture
4-Model Ensemble
💡 What's an ensemble?
Instead of relying on one method, we asked four different "experts" for their prediction, then blended their answers. Think of it like asking a stats nerd, a basketball scout, the selection committee, and a momentum tracker for their picks — then averaging them based on how reliable each expert is. The stats nerd (KenPom) gets the most weight because they're right most often.
KenPom Efficiency
The Stats Nerd — looks at raw scoring efficiency
How many points a team scores vs allows per 100 possessions — the #1 predictor in college basketball
Bradley-Terry
The Scout — weighs quality of wins and losses
Strength ratings computed from who beat whom — beating a good team counts more than beating a bad one
Seed-Based
The Committee — uses the seed numbers
The committee's expert judgment turned into probabilities — a 1 seed beats a 16 seed 99% of the time
Conf Tourney
The Momentum Tracker — hot teams get a bump
Teams that just won their conference tournament get a small confidence boost
132,133
Total Predictions
76.0%
Avg Confidence
12,257
Above 90% Conf
11,015
True Toss-Ups
11,100
Extreme Upsets
/Round of 64
First Round Predictions
These are the actual games happening in the tournament. The longer the bar, the more confident we are.
East Region
We're very confident in this one — Duke wins 96 out of 100 times in our model
Duke
SienaWe genuinely don't know — this is pure March Madness — Ohio St wins 54 out of 100 times in our model
Ohio St
TCUShould win, but don't be shocked if they don't — St John's wins 76 out of 100 times in our model
St John's
Northern IowaWe genuinely don't know — this is pure March Madness — Kansas wins 53 out of 100 times in our model
Kansas
Cal BaptistWe genuinely don't know — this is pure March Madness — South Florida wins 50 out of 100 times in our model
Louisville
South FloridaBarely leaning one way — could go either way — Michigan St wins 58 out of 100 times in our model
Michigan St
N Dakota StShould win, but don't be shocked if they don't — UCLA wins 76 out of 100 times in our model
UCLA
UCFSolid favorite — but upsets happen — Connecticut wins 89 out of 100 times in our model
Connecticut
FurmanSouth Region
We're very confident in this one — Florida wins 96 out of 100 times in our model
Florida
LehighWe genuinely don't know — this is pure March Madness — Iowa wins 52 out of 100 times in our model
Clemson
IowaWe genuinely don't know — this is pure March Madness — Vanderbilt wins 52 out of 100 times in our model
Vanderbilt
McNeese StSolid favorite — but upsets happen — Nebraska wins 81 out of 100 times in our model
Nebraska
TroyWe genuinely don't know — this is pure March Madness — VCU wins 54 out of 100 times in our model
North Carolina
VCUWe're very confident in this one — Illinois wins 93 out of 100 times in our model
Illinois
PennSolid favorite — but upsets happen — St Mary's CA wins 80 out of 100 times in our model
St Mary's CA
Texas A&MWe're very confident in this one — Houston wins 93 out of 100 times in our model
Houston
IdahoWest Region
Should win, but don't be shocked if they don't — Michigan wins 72 out of 100 times in our model
Michigan
HowardShould win, but don't be shocked if they don't — St Louis wins 66 out of 100 times in our model
Georgia
St LouisWe genuinely don't know — this is pure March Madness — Akron wins 54 out of 100 times in our model
Texas Tech
AkronBarely leaning one way — could go either way — Alabama wins 56 out of 100 times in our model
Alabama
HofstraWe genuinely don't know — this is pure March Madness — Tennessee wins 51 out of 100 times in our model
Tennessee
Miami OHSolid favorite — but upsets happen — Virginia wins 84 out of 100 times in our model
Virginia
Wright StBarely leaning one way — could go either way — Santa Clara wins 62 out of 100 times in our model
Kentucky
Santa ClaraSolid favorite — but upsets happen — Iowa St wins 88 out of 100 times in our model
Iowa St
Tennessee StMidwest Region
We're very confident in this one — Arizona wins 93 out of 100 times in our model
Arizona
LIU BrooklynBarely leaning one way — could go either way — Utah St wins 63 out of 100 times in our model
Villanova
Utah StWe genuinely don't know — this is pure March Madness — High Point wins 55 out of 100 times in our model
Wisconsin
High PointShould win, but don't be shocked if they don't — Arkansas wins 74 out of 100 times in our model
Arkansas
HawaiiShould win, but don't be shocked if they don't — BYU wins 68 out of 100 times in our model
BYU
NC StateWe're very confident in this one — Gonzaga wins 96 out of 100 times in our model
Gonzaga
KennesawShould win, but don't be shocked if they don't — Miami FL wins 77 out of 100 times in our model
Miami FL
MissouriSolid favorite — but upsets happen — Purdue wins 89 out of 100 times in our model
Purdue
Queens NC/Model Alert
Upset Picks
💡 What's an upset pick?
In March Madness, the higher seed (like a #1) is supposed to beat the lower seed (like a #16). But our computer found some matchups where the "underdog" is actually the better team statistically. The selection committee may have under-seeded them — our model disagrees with the committee's ranking. These are the games where we're predicting an upset.
KansasKansasAlabama/Elite Matchups
1-Seed vs 1-Seed
💡 Why show these?
The four #1 seeds are supposed to be the best teams in the country. But even among the elite, our model has clear favorites. If Duke played Arizona, who wins? These predictions show which #1 seed our model thinks is actually the strongest — and by how much.
Duke
75% to win
vs Florida (25%)Duke
70% to win
vs Michigan (30%)Duke
64% to win
vs Arizona (36%)Arizona
61% to win
vs Florida (39%)Arizona
59% to win
vs Michigan (41%)Michigan
54% to win
vs Florida (46%)/High Confidence
Most Lopsided
ConnecticutoverSamfordUCLAoverSamfordConnecticutoverSouthern UnivConnecticutoverSF Austin
TexasoverSamfordConnecticutoverCal BaptistConnecticutoverUT San AntonioConnecticutoverMissouri StConnecticutoverHoly CrossConnecticutoverJacksonville/Coin Flip
Biggest Toss-Ups
UC San Diegovs
Holy CrossIowa StvsMiami OHTexas TechvsWisconsinHigh PointvsIowaAlabamavsUtah StNorthern IowavsVillanovaMcNeese StvsWisconsin
VermontvsTexas TechHowardvsClemsonSt Mary's CAvsVirginia/Guide
How to Read This
The Percentages
When we say 75%, imagine the two teams playing 100 games. We think Team A wins about 75 of them. The higher the number, the more confident we are. Even a 90% pick loses 1 out of 10 times — that's why March Madness is so exciting.
The Bar Charts
The colored bars show how confident we are visually. A long teal bar = strong favorite. Two equal-length bars = coin flip. A pink bar on the underdog = we're predicting an upset. The longer the bar, the more confident the pick.
Upset Picks
These are games where our computer disagrees with the "experts" (the selection committee). The committee ranked Team A higher, but our stats say Team B is actually better. This is where the fun is — will our data beat the experts?
How We Get Scored
Kaggle scores us based on how calibrated our predictions are. If we say 70%, roughly 70% of those games should actually go our way. Being confidently wrong gets punished hard — being cautious is safer but won't win the competition.