Finding Nemo: Predicting Movie Performances by Machine Learning Methods

Jong-Min Kim,Leixin Xia,Keon-Hyung Lee,Iksuk Kim,Seungjoo Lee

doi:10.3390/jrfm13050093

Abstract

Analyzing the success of movies has always been a popular research topic in the film industry. Artificial intelligence and machine learning methods in the movie industry have been applied to modeling the financial success of the movie industry. The new contribution of this research combined Bayesian variable selection and machine learning methods for forecasting the return on investment (ROI). We also attempt to compare machine learning methods including the quantile regression model with movie performance data in terms of in-sample and out of sample forecasting.

Highlights

The movie industry has been growing over the several decades which is a global phenomenon
The contribution of our research proposes our method combining Bayesian variable selection and machine learning methods which includes quantile regression when we have an extremely skewed return on investment (ROI) data because there are many films with a low ROI, and some are very successful
0.2772 0.0294 0.00001 0.7174 0.0294 In Table 9, we showed that QR50 has the smallest mean absolute percentage error (MAPE) compared with the other five models (QR25, QR75, multivariate adaptive regression splines (MARS), support vector machine (SVM) and NNet) in terms of ROI

Summary

Introduction

The movie industry has been growing over the several decades which is a global phenomenon. Lu (2019) analyzed qualitative and quantitative analytic hierarchy process method to establish the movie box office prediction model, in combination with the actual data of the Chinese film market. Çaglıyor et al (2019) aimed to design a forecast model using different machine learning algorithms such as support vector regression (SVM), artificial neural networks (ANN), decision tree regression (DT) and linear regression (LR) to estimate the theatrical success of US movies in Turkey before their market entry. With the selected important variables, we analyze and compare quantile regressions, multivariate adaptive regression splines, support vector machine, and neural network methods to form an accurate prediction model of ROI using major film forecasting variables (such as the number of theater screenings, number of running weeks, critics’ reviews, production budget, and genres).

Data and Description

Objective

Empirical Results

Conclusions