March Madness Bracket Predictor

An ensemble of XGBoost and logistic regression models predicting NCAA tournament outcomes using advanced team metrics and historical matchups.

Overview

Built an ensemble prediction model combining XGBoost and stacked logistic regression to predict NCAA tournament game outcomes, achieving 78% accuracy across 10 years of historical matchups. Engineered 40+ differential features from KenPom, EvanMiya, and Barttorvik metrics (adjusted efficiency, ELO, SOS, shooting splits, Four Factors) and applied correlation filtering to reduce multicollinearity.

Tech Stack

R, XGBoost, tidyverse, caret, KenPom, EvanMiya