Predictive Modeling of Movie Revenue
UOC – Applied Statistics Project
Description
As part of an Applied Statistics course, this project aimed to predict the global box office revenue of 3,000 movies from The Movie Database (TMDb). The workflow included exploratory analysis, feature engineering and multiple regression modeling using Python.
Tools
Files
Phase 1: Exploratory Data Analysis
Assessed relationships between revenue and features like budget, runtime, popularity, and language.
Budget showed the highest correlation (~0.75) and was the strongest individual predictor.
Visualized insights using scatter plots and correlation matrices.
Phase 2: Feature Engineering & Regression Modeling
Engineered numerical features from JSON fields, such as number of male cast members and production companies. Built a multiple linear regression model using scikit-learn. Budget and popularity were the most significant predictors. The final model achieved an R² score of ~0.38 and was validated through residual analysis.
Phase 3: Model Validation & Residual Analysis
Evaluated the model’s assumptions by analyzing the distribution of residuals. A histogram revealed a roughly symmetric shape centered around zero, supporting the validity of the linear model for this dataset.
