Project 05
Box Office Revenue Prediction
Classifying films as High or Low box-office revenue relative to their production country's GDP, with logistic regression and decision trees in R, LIME explanations, an R Markdown report and a Shiny app.
Overview
Films released between 2000 and 2024 are classified as High or Low box-office revenue relative to the GDP of their production country. An individual project for UWA's Computational Data Analysis unit in 2025, delivered as an R Markdown report and an interactive Shiny app covering data exploration, feature selection, model training, evaluation and prediction.
My role
Individual project. I did the data preparation, modelling, report and app.
What I did
- Cleaned and joined box-office data with World Bank GDP and US Census population figures, one row per production country.
- Labelled each film High or Low by revenue relative to GDP, split at the median.
- Selected features by information gain and recursive feature elimination, after removing highly correlated predictors.
- Trained logistic regression and decision tree models with caret, using downsampling and 5-fold cross-validation.
- Logistic regression with information-gain features did best: 0.83 accuracy and 0.89 AUC on the test set, with LIME explaining individual predictions.
Problems solved
- GDP appears in both the target and the predictors, and vote counts build up after release, so I documented that the model explains a film's revenue class rather than forecasting it before release.