The bracket model that picked Duke now picks Michigan. The full stacked ensemble says Michigan 66.9%. A late-round specialist model trained only on Elite Eight, Final Four, and championship games says Michigan 78.1%. Ten thousand Monte Carlo simulations were unanimous. The only catch: UConn has already beaten this model’s picks twice.
The machine learning bracket model went 5-for-8 in the Sweet 16 and 2-for-4 in the Elite Eight. Then Duke’s championship prediction died on Braylon Mullins’ three-pointer with under a second left. The Final Four is set: UConn vs. Illinois, Arizona vs. Michigan.
The model went 33 for 48 through two rounds. Five Round of 32 upsets blew up half the bracket. Here’s what the stacked ensemble says about all eight Sweet 16 games.