Machine Learning Applications on Box-Office Revenue Forecasting: The Taiwanese Film Market Case Study

Abstract

The Random Forest algorithm (RFA) is used to predict the approximate final box-office revenue of a movie in the Taiwanese film market. The results show that the RFA has stable capabilities to predict the final box-office revenue of a movie during its theatrical period with an 80% overall accuracy. Two other machine learning algorithms, i.e., the Support Vector Machine and the Logistic Regression algorithms, are applied for comparison with the RFA. We find that the RFA still achieves the highest overall accuracy of prediction in our experiment. Additionally, we applied an unsupervised machine learning method to distinguish each group in the box office revenue categories in the classification problem. Also, the feature importance analysis indicates that word-of-mouth plays a vital role in theatrical revenue determination. Our findings imply several crucial suggestions for film distributors.

Appendices

Appendix

Self-Organizing Maps Algorithm Pseudocode for Box Office Revenue in Taiwan

Self-Organizing Maps Clustering Results and Illustration of Training Error

Case 1: Topology: 10*10, neighborhood radius: 10 $\sqrt{2}$, shrink learning rate: 0.9; iteration: 100.

Case 2: Topology: 10*10, neighborhood radius: 10 $\sqrt{2}$, shrink learning rate: 0.01; iteration: 100.

Case 3: Topology: 8*8, neighborhood radius: 8 $\sqrt{2}$, shrink learning rate: 0.9; iteration: 100.

Case 4: Topology: 8*8, neighborhood radius: 8 $\sqrt{2}$, shrink learning rate: 0.01; iteration: 100.