And the Oscar Goes to... (An Application and Comparison of Models Used to Predict the Winner of the Academy Award for Best Picture)
Open Access
Author:
Slayton, Joshua
Area of Honors:
Statistics
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Michael Rutter, Thesis Supervisor Michael Rutter, Thesis Honors Advisor Terry Mark Blakney, Faculty Reader
Keywords:
statistics logistic regression decision tree leave-one-out cross validation oscars academy awards best picture prediction data analysis classification regression
Abstract:
Each year, the Academy of Motion Picture Arts and Sciences recognizes exceptional achievements in cinema with its Oscars ceremony. Of all the awards handed out at this annual celebration, the Academy Award for Best Picture is arguably the most coveted. This project involves using historical Oscars data, as well as data from other awards shows such as the Golden Globes, to develop and compare models for predicting the winners of the Academy Award for Best Picture. In particular, models using both logistic regression and decision tree classification will be developed. Model performance will be evaluated using the leave-one-out cross validation procedure to compute prediction accuracies and root mean square errors for each model. These measures will allow direct comparisons between the models to be made, which should lead to some interesting results. For example, models that incorporate more than just Oscar nominations data perform better, unsurprisingly, than those that do not. Also, as will be discussed, the decision tree classification models all have higher prediction accuracies than their corresponding logistic regression models. The predictions that the “best” logistic regression and decision tree classification models got most incorrect will be explored. Finally, the models generated through this project will be applied to make predictions for this year’s Best Picture winner, to be announced on April 25, 2021.
