In this project, we focus on predicting the winners of esports tournaments, specifically for Super Smash Bros. Melee. While this is an older game released in 2001, it still sees a vibrant esports scene, with tournaments regularly offering tens of thousands of dollars in prize money, and viewership sometimes in the hundreds of thousands.

Our goal was the same as any other data scientist analyzing sports data - predict the winner as accurately as possible. In particular, we wanted to develop a machine learning model that could predict the outcome of individual matches between players, and develop a second model that could predict the winner out of the top 8 finalists in a tournament. We placed a heavy emphasis on engineering predictive features beyond standard Elo scores.

Ultimately, our single-match predictor was able to outperform the baseline of "predicting whoever has the highest elo", with a statistically significant improvement in accuracy.