Towards an Effective Over-The-Top Platform Service: A Machine Learning Approach for Box Office Analysis

We developed a machine learning approach for trend analysis of box office performance on online platforms. We first collected the top 300 box office data points, including movie name, release date, production country, genre, director, actor, and screening time from the Korean Film Council online integrated computer network. We preprocessed these data to conFigure a total of 1030 input variables and then used a deep autoencoder and uniform manifold approximation and projection to reduce them to two input variables. Next, we used a k-means clustering algorithm to cluster movies with similar characteristics on the reduced data. We constructed an extreme gradient boosting (XGBoost) model using the 1030 original variables and cluster results as input and output variables for each cluster. We finally demonstrated the interpretability of the XGBoost model using Shapley additive explanations to understand the clustering results. The proposed method is expected to assist in the marketing strategy of domestic entertainment companies because it can be used for casting for movie production, determining release time, and adjusting running time by determining elements that determine box office hits.