Abstract
Title: Madness in March: Leveraging AI and Statistical Analysis to Predict NCAA Tournament Outcomes Better than Humans
Authors: Beckett, N. 1,2
, Randall, W. 1
, Montney, J. 1
, Hull, K. 2 , Gannon, D. 1
Affiliations: 1Department of Health and Human Performance, Fort Hays State University 2Department of Athletics, Fort Hays State University
Abstract: The NCAA Men's Basketball Tournament, commonly known as "March Madness," operates as a high-variance system where traditional seeding frequently fails as a predictive metric. Millions of brackets are submitted annually, yet high-accuracy forecasting remains elusive due to human bias and subjective guidelines. This study evaluates the effectiveness of an artificial intelligence generated, 18-metric equation designed to strip away traditional seeding bias and predict tournament game outcomes more accurately than human consensus. Utilizing Google AI Studio’s 3.1 Pro SOTA reasoning model, an initial feature selection process identified 18 advanced statistical variables that historically correlate tightly with tournament success. The AI filtered out "noisy" volume statistics, such as raw points per game, and assigned dynamic mathematical weights to the selected variables. This targeted equation was then applied to evaluate bracket matchups and compute a confidence score (Z-score) for each prediction, explicitly seeking to identify statistically undervalued teams primed for tournament runs. The AI-driven protocol yielded exceptional predictive results, achieving a 74.6% overall accuracy rate. Most notably, the model exhibited a 0% false-positive rate on upset predictions; when the equation predicted a lower seed to win, it was correct 100% of the time (6/6). Furthermore, the model recorded 100% accuracy (22 of 22) on statistically confident Round of 64 matchups. Ultimately, the AI bracket placed in the 78th percentile among 26.6 million ESPN submissions. These findings validate the hypothesis that the "chaos" of March Madness contains highly quantifiable indicators, demonstrating that an objective, mathematically weighted AI model can successfully isolate true team value and consistently outperform traditional human intuition.
Department/Program
Health and Human Performance
Submission Type
in-person poster
Date
4-8-2026
Rights
Copyright the Author(s)
Recommended Citation
Beckett, Noah; Randall, Whitney; Montney, Justin L.; Gannon, Andrew; and Hull, Kyle
(2026)
"Madness in March: Leveraging AI and Statistical Analysis to Predict NCAA Tournament Outcomes Better than Humans,"
SACAD: Scholarly Activities: Vol. 2026, Article 40.
Available at:
https://scholars.fhsu.edu/sacad/vol2026/iss2026/40
