Movie Actor Success Prediction

Open Access
Wagura, David W
Area of Honors:
Interdisciplinary in Economics and Industrial Engineering
Bachelor of Science
Document Type:
Thesis Supervisors:
  • Soundar R T Kumara, Thesis Supervisor
  • Catherine Mary Harmonosky, Honors Advisor
  • James R. Tybout, Honors Advisor
  • Movies
  • Actors
  • Prediction
  • Film
  • Analysis
  • Networks
  • Mathematical Program
The movie industry is large with billions of dollars in revenues raised each year. To create movies, studios must consider several different combinations of variables to determine which ones will make the most successful films. Variables like genre, budget, actors and others must be considered and there are many different possibilities that studios can use. This thesis looks specifically at the actor and technical personnel selection to predict a movie’s success. The assumption being that once a studio gets to the point of looking for actors, they already know some of the main attributes like genre, potential rating and budget. The main research question addressed in this thesis is about the considerations needed to select an actor or combinations of actors that can result in a successful movie. This was done by first analyzing film data to find trends in the data set. The analysis determined that most attributes have bi-modal distributions that point to distinct groups within the data set. From this conclusion, the films were clustered and six clusters were found in the data set. It was then postulated that studios tend to have multiple objectives in filmmaking and due to data availability, the cluster that focused on financial success was chosen for further analysis. A metric called, earnings per theatrical engagements (ER), was defined and used to describe both the films and the actors. Several options were then evaluated to predict the best actors. The first option was a mathematical program that uses an objective function to solve for an actor or actors that maximize ER or other measures of success. If only one actor is selected, it was shown that one could then build a network to find the best combinations of actors. Using the data from the cluster it was shown that a search method could be used where a database is queried to find the best actor. A few examples were given where well know actors like Leonardo DiCaprio and Daniel Radcliffe were found by the query. It was then shown how a network could be used to find actors that would work best with the initial actor. Finally, a few ideas are presented as to how the mathematical program or the network can be used to predict the likelihood of success of a new actor