Predicting outcomes of mixed martial arts fights with novel fight variables

In this study, I attempt to forecast the win/loss outcomes of mixed martial arts bouts with fight data. Both basic count variables and newer, constructed variables are considered. These novel measures are then used to predict wins and losses using a linear regression model, and this model is compared to baseline models in terms of predictive ability. The final model contains both count variables and constructed variables and is found to have significantly greater predictive ability than baseline models. Cross-validation and discriminant analysis confirm these results.