Predicting Madness: The Most Likely March Madness Upsets by Region
Most March Madness models predict matchup results with approximately 75% accuracy. By leveraging advanced analytics on team offensive efficiency, defensive efficiency, personnel, and box scores, I developed two models—Logistic Regression and Random Forest—that achieve around 81% accuracy in predicting the first round of March Madness.
Both models prioritize overall offensive and defensive efficiency but differ in the specific factors they emphasize. The logistic regression model favors teams that excel in defensive rebounding and free-throw shooting while being wary of those reliant on mid-range and three-point shot-making. In contrast, the random forest model values teams with strong perimeter defense and low turnover rates.
I have used these two models to compile the top 2-3 most likely first round upsets from each region.
Bold = Upset Predicted
South Region
(12) UC San Diego vs (5) Michigan
My logistic regression model predicts UC San Diego to pull off the upset, giving them a 57% chance to win, while my random forest model puts their chances at 49%. UC San Diego excels defensively, boasting a highly favorable turnover differential—forcing 7.3 more turnovers per game than they commit—compared to Michigan, who gives up 2.7 more turnovers than they generate. Additionally, UC San Diego’s guards and wings provide strong support on the defensive glass, securing about 8 more defensive rebounds per game than Michigan’s perimeter players. The turnover and rebounding difference could lead UC San Diego to the upset.
(13) Yale vs (4) Texas A&M
While the logistic regression model isn’t high on Yale’s chances in this matchup, the random forest model gives them a 44% chance to win. This is largely due to Yale's superior shooting efficiency—shooting over 5% better from the field, more than 7% better from three, and over 4% better from the free-throw line, resulting in a 7% higher effective field goal percentage (eFG%). Yale has six players who attempt more than 1.5 three-pointers per game while shooting over 38%, including John Poulakidas, who takes 7.8 threes per game at a 41% clip. They also commit 2.3 fewer turnovers per game than Texas A&M. However, as mentioned earlier, the logistic regression model is less favorable toward teams that rely on jump shooting and instead values Texas A&M’s strength on the offensive glass and defensive efficiency. If Yale has a hot shooting night, don’t be surprised if they pull off the upset.
West Region
(12) Colorado St. vs (5) Memphis
Both the logistic regression and random forest models project a Colorado State victory, averaging a 73% win probability. Colorado State boasts a deep roster, with its bench contributing over 7.5 more points per game than Memphis’ reserves. They also have a more efficient offense, generating higher-percentage looks at the rim and the free-throw line, leaving Memphis’ primary scoring edge in three-point shooting. Defensively, Colorado State excels both on the perimeter and in the paint. While Memphis is a strong perimeter defensive team, Colorado’s limited reliance on three-point shooting may neutralize that advantage.
(11) Drake vs (6) Missouri
Both models give Drake about a 47% chance of winning this first-round matchup. While Missouri boasts the more efficient offense, Drake’s best path to victory lies in dominating the interior. Over the season, their centers have combined to average nine more points and seven more rebounds per game than Missouri’s. However, this will be a challenge against a Missouri defense that has held opponents to under 49% shooting from two-point range. Another key factor for Drake is defensive efficiency—they allow the fewest points per game in the country and have limited opponents to just 31% shooting from three. This could be crucial against a Missouri team that shoots 38% from beyond the arc, ranking in the top 20 nationally.
(9) Oklahoma vs (8) UConn
The logistic regression model gives Oklahoma a 62% chance of winning, while the random forest model puts them at 37%. However, I kept them as a No. 3 seed due to the traditional 9 vs. 8 seeding. Oklahoma’s path to victory hinges on two key factors: elite defense and getting to the free-throw line. They rank in the top 12 nationally in both free throws attempted and free throw percentage, generating 14 points per game at the line. Meanwhile, they hold opponents to just 15 free-throw attempts per game—a mark UConn hasn’t even averaged this season. The biggest concern for Oklahoma is the matchup in the paint, where UConn’s centers have a significant edge, averaging six more points and six more defensive rebounds per game than Oklahoma’s.
East Region
(11) VCU vs (6) BYU
While the logistic regression model gives VCU a 40% chance of winning, the random forest model favors them at 64%. This matchup is a classic clash between BYU’s high-powered offense and VCU’s elite defense. BYU excels in both volume and efficiency, ranking 7th in two-point percentage and 25th in field goal percentage. However, VCU counters with stifling defense, allowing the 3rd lowest two-point percentage, 8th lowest field goal percentage, and ranking in the top 20 for three-point defense. For VCU to secure the win, they’ll need to contain BYU’s dominant inside scorer and rebounder, Keba Keita, as well as the elite shooting duo of Richie Saunders and Trevin Knell.
(12) Liberty vs (5) Oregon
Once again, the logistic regression model is skeptical of Liberty’s chances, while the random forest model predicts them to win. This continues the trend of the random forest model favoring elite shooting teams—Liberty ranks among the top six shooting teams in the country, significantly outperforming Oregon in efficiency. Liberty also boasts four high-volume three-point shooters, all hitting at 38% or better. For Liberty to overcome Oregon’s depth and size advantage, those shooters will need to be lights out from beyond the arc.
(9) Baylor vs (8) Mississippi St.
The logistic regression model gives Baylor a 57% chance of winning, but I ranked them lower since it’s a close 9 vs. 8 matchup. This prediction is driven by Mississippi State’s defensive struggles, as they have difficulty defending both the interior and perimeter while frequently sending opponents to the free-throw line—an area where Baylor thrives. To overcome their defensive issues, Mississippi State will need to lean on their inside scoring to keep pace.
Midwest Region
(13) High Point vs (4) Purdue
Neither model predicts a full upset in the Midwest region, but High Point’s 41% win probability—given by the random forest model—is the highest of any lower seed in this region. High Point thrives on the glass, excelling in both offensive and defensive rebounding, particularly in creating second-chance scoring opportunities. Additionally, their bench has contributed nearly eight more points per game than Purdue’s. To pull off the upset, High Point will need to capitalize on the minutes when Braden Smith and/or Trey Kaufman-Renn are off the floor.
