Investigation of time series-based genre popularity features for box office success prediction

Predicting the profitability of movies at the early phase of production can be helpful to support the decision to invest in movies however, due to the limited information at this stage it is a challenging task to predict the movie's profitability. This study proposes genre popularity features using time series prediction. We argue that a movie can produce better box office returns if its genre's popularity is high at the time of release. The novel genre popularity features are proposed in terms of budget, revenue, frequency, success, and return on investment (ROI). The proposed features couple the predicted genre popularity with release time, in order to train the machine learning classifiers. The experimentation shows that the Gradient Boosting classifier gained a significant improvement using proposed features and achieved an accuracy of more than 92.4%, i.e., 35.7% better than an existing state of the art study considering a multi-class problem.