Predicting winners in the NCAA Basketball Tournament

Abstract

Every year, the NCAA basketball tournament, known as “March Madness,” has 68 of the best college basketball teams in the country face off in a single-elimination tournament to crown a champion. Significant work has been devoted to the development of models to accurately predict the winner in any head-to-head matchup. In 2014, the website Kaggle created a competition that had teams of people trying to create a model and accurately predict the 2014 tournament. Most submissions had a higher accuracy rate than the average person, but nobody came close to a perfect bracket. In this report, I use data from the 2014 Kaggle competition and recent advances in statistical methodology to determine the best models for predicting winners of head-to-head matchups. I then compare my results to those from the best submissions from the Kaggle competition.