The commercial success of a movie remains difficult to anticipate during early production stages, despite the availability of digital information on audience engagement and film attributes. While prior studies have explored box office or rating prediction, few have combined early-stage feature sets with transparent, interpretable models. This study addresses this gap by developing a large-scale prediction framework using machine learning and explainable AI techniques.
The analysis draws on a dataset of 433,350 films from IMDb and TMDB. Multiple predictive models were employed alongside SHAP and LIME to identify the factors most strongly associated with success, defined through a dual rating–vote threshold.
The results show that popularity metrics, director and thematic attributes are the strongest predictors of success, while genre effects such as comedy and drama play a more modest but still meaningful contextual role.
The study contributes a scalable, interpretable, early-stage prediction approach that supports producers, investors and distributors by highlighting both the drivers of success and the reasoning behind model outputs.
