How do you pick a perfect March Madness bracket. Guessing? Doubtful. Extreme luck? Almost certainly. Machine learning? Let's find out...

​

I fed a Lasso Logistic Regression (lasso) model KenPom.com statistics such as offensive efficiency, defensive efficiency, and tempo for tournament teams since 2002. I also gave the model previous outcomes for each tournament game. The lasso takes training data, figures out what variables are important, and then predicts new outcomes. 

​

Does it work? Not really. In the years I have tested it, I've produced a bracket on par with picking chalk. It often finds some underrated teams in the first round, but rarely the big upsets. Still, it's a fun way to pick a bracket or a single matchup. I deployed the model as a Shiny app (attached to this page) so anyone can use it. Check back each March for a new version!