Forecasting U.S. Movie Gross Revenues : A Random Forest Classifier Approach Based on Pre-production Data

Abstract: Blockbusters screening in cinemas often include star-studded casts in hope of becoming profitable. However, producing a movie is associated with great risk. More than half of the movies released between 2008 and 2012 in the U.S. failed to turn a profit and the top 10% of movies accounted for nearly 70% of the box-office revenue in 2012. Estimating the revenue for a movie prior to the release is therefore of great importance for decision makers in the movie industry. Large movie production companies decide on greenlighting movies years in advance of their releases, ruling out a need for prediction models based on features available only near the release. In this report we present a random forest classifier for predicting the revenue of movies produced in the U.S. based solely on pre-production factors. We design a novel method of calculating the prestige of actors based on revenue generated by their previous movies. When forecasting the revenue of a movie into nine different revenue range classes, our model demonstrated an accuracy of 29% for exact class predictions and 59% when allowing for consideration of adjacent classes. We also found no strong correlation between revenue generated by actors’ previous movies and their ability to shape revenue in future movies.