Machine Learning Models for Revenue Prediction in the Film Industry: A Comparative Study

Abstract

This thesis addresses the challenge of accurately predicting movie revenue, a critical task for distribution companies and stakeholders in the film industry. The study focuses on constructing and analyzing a dataset comprising movie revenue data from predominantly 2019 to 2023, encompassing diverse genres, production budgets, and talents. The primary objective is to develop machine learning models capable of forecasting initial weekend revenues and subsequent revenue trends for three target movies: "Dune: Part Two," "Kung Fu Panda 4," and "Ghostbusters: Frozen Empire." The methodology involves leveraging machine learning algorithms, specifically Gradient Boosting and Random Forests, to train predictive models using filtered subsets of the dataset based on genres, directors, and production companies. Results indicate varying degrees of predictive accuracy, with the Gradient Boosting algorithms demonstrating the highest efficacy in revenue prediction. However, limitations such as selection bias in the dataset, challenges in model interpretability, and difficulties in predicting flopping outliers highlight areas for future research and model refinement. Despite these limitations, the study underscores the potential of machine learning in enhancing revenue forecasting in the film industry and could provide insights for industry practitioners and decision-makers.